S.Hrg.119-143

1[Senate Hearing 119-143]2[From the U.S. Government Publishing Office]34                                                        S. Hrg. 119-14356                 WINNING THE AI RACE: STRENGTHENING U.S.7                CAPABILITIES IN COMPUTING AND INNOVATION89=======================================================================1011                                HEARING1213                               before the1415                         COMMITTEE ON COMMERCE,16                      SCIENCE, AND TRANSPORTATION17                          UNITED STATES SENATE1819                    ONE HUNDRED NINETEENTH CONGRESS2021                             FIRST SESSION2223                               __________2425                              MAY 8, 20252627                               __________2829    Printed for the use of the Committee on Commerce, Science, and30                             Transportation3132                 [GRAPHIC NOT AVAILABLE IN TIFF FORMAT]3334                Available online: http://www.govinfo.gov3536                               ______3738                 U.S. GOVERNMENT PUBLISHING OFFICE394061-426 PDF                WASHINGTON : 20254142       SENATE COMMITTEE ON COMMERCE, SCIENCE, AND TRANSPORTATION4344                    ONE HUNDRED NINETEENTH CONGRESS4546                             FIRST SESSION4748                       TED CRUZ, Texas, Chairman4950JOHN THUNE, South Dakota             MARIA CANTWELL, Washington,51ROGER WICKER, Mississippi                Ranking52DEB FISCHER, Nebraska                AMY KLOBUCHAR, Minnesota53JERRY MORAN, Kansas                  BRIAN SCHATZ, Hawaii54DAN SULLIVAN, Alaska                 EDWARD MARKEY, Massachusetts55MARSHA BLACKBURN, Tennessee          GARY PETERS, Michigan56TODD YOUNG, Indiana                  TAMMY BALDWIN, Wisconsin57TED BUDD, North Carolina             TAMMY DUCKWORTH, Illinois58ERIC SCHMITT, Missouri               JACKY ROSEN, Nevada59JOHN CURTIS, Utah                    BEN RAY LUJAN, New Mexico60BERNIE MORENO, Ohio                  JOHN HICKENLOOPER, Colorado61TIM SHEEHY, Montana                  JOHN FETTERMAN, Pennsylvania62SHELLEY MOORE CAPITO, West Virginia  ANDY KIM, New Jersey63CYNTHIA LUMMIS, Wyoming              LISA BLUNT ROCHESTER, Delaware6465                 Brad Grantz, Republican Staff Director66           Nicole Christus, Republican Deputy Staff Director67                     Liam McKenna, General Counsel68                   Lila Harper Helms, Staff Director69                 Melissa Porter, Deputy Staff Director70                     Jonathan Hale, General Counsel7172                            C O N T E N T S7374                              ----------75                                                                   Page76Hearing held on May 8, 2025......................................     177Statement of Senator Cruz........................................     178Statement of Senator Cantwell....................................     379Statement of Senator Sheehy......................................    2980Statement of Senator Moreno......................................    3381Statement of Senator Klobuchar...................................    3582Statement of Senator Schatz......................................    4083Statement of Senator Budd........................................    4284Statement of Senator Kim.........................................    4485Statement of Senator Schmitt.....................................    4686Statement of Senator Hickenlooper................................    4887Statement of Senator Curtis......................................    5088Statement of Senator Duckworth...................................    5389Statement of Senator Young.......................................    5590Statement of Senator Blunt Rochester.............................    5691Statement of Senator Moran.......................................    5892Statement of Senator Lujan.......................................    6093Statement of Senator Lummis......................................    6394Statement of Senator Rosen.......................................    6495Statement of Senator Sullivan....................................    6696Statement of Senator Markey......................................    6897Statement of Senator Peters......................................    7198Statement of Senator Fetterman...................................    7399100                               Witnesses101102Sam Altman, Co-Founder and Chief Executive Officer, OpenAI.......     5103    Prepared statement...........................................     7104Dr. Lisa Su, Chief Executive Officer and Chair, Advanced Micro105  Devices (AMD)..................................................     9106    Prepared statement...........................................    10107Michael Intrator, Co-Founder and Chief Executive Officer,108  CoreWeave......................................................    12109    Prepared statement...........................................    14110Brad Smith, Vice Chair and President, Microsoft Corporation......    21111    Prepared statement...........................................    22112113                                Appendix114115Response to written questions submitted to Sam Altman by:116    Hon. Roger Wicker............................................    83117    Hon. Marsha Blackburn........................................    83118    Hon. Maria Cantwell..........................................    84119    Hon. Amy Klobuchar...........................................    86120    Hon. Brian Schatz............................................    86121    Hon. Edward Markey...........................................    89122    Hon. Tammy Baldwin...........................................    91123    Hon. Jacky Rosen.............................................    91124    Hon. Lisa Blunt Rochester....................................    92125Response to written questions submitted to Dr. Lisa Su by:126    Hon. Todd Young..............................................    92127    Hon. Maria Cantwell..........................................    93128    Hon. Amy Klobuchar...........................................    94129    Hon. Brian Schatz............................................    95130    Hon. Jacky Rosen.............................................    96131    Hon. John Fetterman..........................................    97132Response to written questions submitted to Michael Intrator by:133    Hon. Maria Cantwell..........................................    99134    Hon. Amy Klobuchar...........................................   100135    Hon. Brian Schatz............................................   101136    Hon. Edward Markey...........................................   102137    Hon. Jacky Rosen.............................................   103138Response to written questions submitted to Brad Smith by:139    Hon. Roger Wicker............................................   104140    Hon. Todd Young..............................................   104141    Hon. Maria Cantwell..........................................   106142    Hon. Brian Schatz............................................   109143    Hon. Edward Markey...........................................   111144    Hon. Tammy Baldwin...........................................   114145    Hon. Jacky Rosen.............................................   115146    Hon. John Fetterman..........................................   117147    Hon. Lisa Blunt Rochester....................................   118148149                WINNING THE AI RACE: STRENGTHENING U.S.150               CAPABILITIES IN COMPUTING AND INNOVATION151                     IN COMPUTING AND INNOVATION152153                              ----------154155                         THURSDAY, MAY 8, 2025156157                                       U.S. Senate,158        Committee on Commerce, Science, and Transportation,159                                                    Washington, DC.160    The Committee met, pursuant to notice, at 10:02 a.m., in161room SR-253, Russell Senate Office Building, Hon. Ted Cruz,162Chairman of the Committee, presiding.163    Present: Senators Cruz [presiding], Moran, Sullivan, Young,164Budd, Schmitt, Curtis, Moreno, Sheehy, Lummis, Cantwell,165Klobuchar, Schatz, Markey, Peters, Duckworth, Rosen, Lujan,166Hickenlooper, Fetterman, Kim, and Blunt Rochester.167168              OPENING STATEMENT OF HON. TED CRUZ,169                    U.S. SENATOR FROM TEXAS170171    The Chairman. Good morning. The Senate Committee on172Commerce, Science, and Transportation is called to order.173    Welcome to our witnesses. Thank you for joining us this174morning.175    In the last two years AI has brought the United States and176the world to a critical inflection point. AI may be a177technology as transformative as the Internet or even more so.178    It has unleashed a new global industrial revolution with179the potential to unlock opportunities that improve our quality180of life, create jobs, and stimulate economic growth.181    The country that leads in AI will shape the 21st century182global order. As a matter of economic security, as a matter of183national security, America has to beat China in the AI race.184    China has made AI central to its national strategy and185China aims to lead the world in AI by 2030, investing heavily186in AI adoption across industries like manufacturing and187defense.188    In this race the United States is facing a fork in the189road. Do we go down the path that embraces our history of190entrepreneurial freedom and technological innovation or do we191adopt the command and control policies of Europe?192    I would suggest that Congress draw on the lessons we can193learn from the dawn of the internet. In the early 1990s194Washington embraced the Internet and explicitly adopted a style195of regulation that was intentionally and decisively a light196touch.197    Congress chose to deregulate under the Telecommunications198Act of 1996 while President Clinton pursued tariff agreements199and treaties that protected America's intellectual property and200technological exports.201    Further, in 1998 Congress enacted a 10-year Internet tax202moratorium so that state laws would not balkanize and stymie203the promise of e-commerce.204    The results of these decisions were extraordinary. By 2000,205the United States has recorded five straight years of historic206highs in productivity gains and investment growth.207    Hundreds of thousands of new jobs were created and the208United States became a top tech exporter with massive sums of209private investment pouring into the U.S. digital economy.210    By contrast, EU countries pursued a series of heavy-handed211regulations that proved enormously costly. In 1993 the United212States and Europe had economies virtually identical in size.213    Today, the American economy is more than 50 percent larger214than Europe's. The drivers of that are tech and the shale215revolution. Those two comprise virtually the entirety of that216massive growth over Europe.217    According to one EU Commission report only 6 percent of218global AI startup funding flows to EU firms--6 percent. That is219one-tenth of the amount that is going to American companies.220    The report directly blames this yawning chasm on the EU's221nasty regulatory approach. And, yet, the Biden administration222for inexplicable reasons tried to align AI policy with the EU223to adopt their failed policies.224    President Biden's sweeping AI executive order, the longest225executive order in American history, cast AI as dangerous and226opaque, laying the groundwork for audits, for risk assessments,227and regulatory certifications.228    Biden's approach inspired similar efforts in state229legislatures across the country, threatening to burden230startups, developers, and AI users with heavy compliance costs.231    Some of my colleagues suggest that a friendlier version of232the Biden approach makes sense. They want a testing regime to233guard against AI, quote, ``discrimination'' and have government234agents provide, quote, ``guidance documents,'' seemingly235something out of Orwell, that will usher in what they call best236practices, as if AI engineers lack the intelligence to237responsibly build AI without the bureaucrats.238    Many in the industry foolishly have supported such239paternalism. Harmful regulations take many forms. Biden's240misguided midnight AI diffusion rule on chips and model weights241would have crippled American tech companies' ability to sell AI242to the world.243    The Biden plan would have handed over key markets to China.244We should want foreign countries, particularly our allies, to245buy American.246    I vocally opposed this rule for months and, indeed, the247Ranking Member and I together urged the Biden administration248not to adopt it and I am very pleased that President Trump has249now confirmed he plans to rescind it.250    All of this busybody bureaucracy, whether Biden's251industrial policy on chip exports or industry and regulator-252approved guidance documents, is a wolf in sheep's clothing.253    To lead in AI the United States cannot allow regulation,254even the supposedly benign kind, to choke innovation or255adoption.256    American dominance in AI depends on two factors: innovation257and adoption. Innovation drives breakthroughs in global258competitiveness. Adoption ensures that these tools empower259American workers and businesses, enabling the United States to260become the world's leading adopter and exporter of AI.261    Thankfully, President Trump has, largely, reversed Biden's262misguided AI agenda. In fact, I think AI was a sleeper issue in263this last election.264    Americans wanted to see President Trump and Republicans265and, indeed, all senators champion AI policies focused on266innovation and adoption.267    The contrast has been astounding. This year, there have268been over $1 trillion of new AI projects including major269investments in Texas like the CoreWeave data center in Plano270and the $500 billion project Stargate in Abilene by OpenAI and271Oracle and others.272    Adopting a light touch regulatory style for AI will require273Congress to work alongside the President just as Congress did274with President Clinton. We need to advance legislation that275promotes long-term AI growth and innovation.276    That is why I will soon release a new bill that creates a277regulatory sandbox for AI modeled on the approach taken by278Congress and President Clinton at the dawn of the Internet that279will remove barriers to AI adoption, prevent needless state280overregulation, and allow the AI supply chain to rapidly grow281here in the United States.282    That is how we will accelerate economic growth, secure U.S.283dominance in AI, and beat China.284    And with that, I turn to Ranking Member Cantwell.285286               STATEMENT OF HON. MARIA CANTWELL,287                  U.S. SENATOR FROM WASHINGTON288289    Senator Cantwell. Thank you, Mr. Chairman.290    Thank you for this hearing, and welcome to the witnesses291before us--Mr. Altman, Dr. Su, Mr. Intrator, and Mr. Smith.292    It is a great pleasure to have all of you here, but it is293an especially prideful moment for the Pacific Northwest to have294Mr. Smith and Mr. Altman here, both representing an OpenAI295approach.296    By that I mean an approach where we want to win against297China and a closed system by making sure that what is developed298here in the United States and around the globe is an299architecture where the United States wins and is open.300    To do that we need to focus in winning on computing power,301on algorithms, and on robust data sources. All of that will be302key.303    Personally, I believe a continued investment in NSF helps304in all of those areas as a good public-private partnership with305the industry that is represented here today.306    I am so proud that we passed the CHIPS and Science Act,307because the CHIPS and Science Act also set a foundation for308investing in the United States of America and bringing more of309the supply chain back to the United States of America to build310on a future leadership that we already have, I believe, in the311computing power.312    But we also need to understand that we have to move forward313on the CHIPS Act like the University of Washington $10 million314grant on multi-design sets for chips, the very large-scale315integrated designs I am sure that Dr. Su will tell us about316today.317    But the fact that the United States has to continue to lead318on the future designs and the implementation of that also319requires us to be very smart about data centers, about sources320of electricity, and how we are going to build that supply that321could be up to 12 percent of electricity demand in the very322near future.323    So how do we do that? I have noticed in each of your324testimonies you all explain this, but I am also very proud that325Microsoft has already signed an agreement with one company,326Fusion Energy in Everett, Washington, for a power source327supply--maybe Mr. Altman in his testimony will talk about328this--but that they hope to get very near future energy source329from that.330    So, clearly, the United States leading on electricity and331development so, Mr. Smith, very much appreciate in your332testimony the accentuation on the fact that the United States333of America needs hundreds of thousands of new electricians,334something we should all want to get behind.335    The fact that having electricity and the electricians and336the data source centers here in the United States and in other337places will be key.338    While I want to see us move forward, as the Chairman said--339we signed a letter saying we needed a broader support for340export controls--I want to be clear.341    Export controls are not a trade strategy. They are not a342back pocket issue that the President of the United States whips343out in trade negotiations.344    We are going to move fast because we are going to set345standards. I believe those standards should be encouraging very346broad distribution of U.S.-manufactured and -made AI chips and347technology, and that we are asking our partners overseas to348comply with the rules that we establish, things like making349sure that there is no circumvention of the supply that somehow350gets into China's hands, making sure that we have access, and351making sure that we can verify on that, and also making sure352that U.S. data companies and cloud-based companies are allowed353to be in that market.354    We should not be going to markets overseas only to have355them tell us that organizations with cloud services from the356U.S. would not be allowed. This, I believe, would be a robust357initiative on getting U.S. AI chips and U.S. AI open systems358dominated around the globe.359    Why do we need to move fast? We need to move fast because360if we do not we are looking at another Huawei, another instance361where the United States is behind and also saying we should362tear out this system that now we do not like for lots of363reasons and back door policies.364    So I am all for winning. That is why we passed the CHIPS365and Science Act. I am all for winning and that is why we have366passed seven bills out of this committee last year that, kind367of, got stuck in the lame duck.368    I think the Chairman of the Committee was not ready to move369forward in negotiations with the House and Senate on those370seven bills. But those bills, a bill between myself and Senator371Young on the Institute for Standards--NIST standards--which I372think we still need to do.373    My colleague and I--Moran--on education and scholarships,374small business, and the bill by my colleague here, Senators375Klobuchar and Thune, which was also related to the NIST376standards.377    So we had an opportunity a year ago to move fast. We did378not do it. So let us do this now. Let us get together and379figure this out. The faster the United States moves now, I like380this great Paul Romer quote, which was about collaboration is381the next phase of innovation.382    If we do not collaborate here, if we throw down on politics383instead of getting the policy right, we will not move fast. Let384us allow these people to do what they do best and let us make385sure the United States has the right policies in place so that386our OpenAI standard wins the day.387    Thank you, Mr. Chairman.388    The Chairman. Thank you.389    I would now like to introduce our witnesses for today. Each390of our witnesses and their companies represent critical parts391of the AI infrastructure, hardware, and software supply chain.392    Our first witness is Sam Altman, the Co-Founder and CEO of393OpenAI. OpenAI is one of the world's most advanced AI394companies, known best for its ChatGPT product.395    Our second witness is Lisa Su, the Chair and CEO of396Advanced Micro Devices--AMD. AMD develops high-performance397processors, graphic chips, and AI accelerators that power398artificial intelligence, and Dr. Su is also a Texan.399    Our third witness today is Michael Intrator, the CEO and400Co-Founder of CoreWeave, an AI hyperscaler. CoreWeave is the401world's largest purpose built AI cloud platform.402    And our final witness is Brad Smith, the Vice Chair and403President of Microsoft. I believe everyone is familiar with his404company.405    Mr. Altman, you are recognized for your opening statement.406If you could turn on the volume.407    Mr. Altman. Sorry about that.408    The Chairman. And I do enjoy telling techies how to operate409the tech.410    Mr. Altman. It is pretty embarrassing that I could not411figure that out.412413    STATEMENT OF SAM ALTMAN, CO-FOUNDER AND CHIEF EXECUTIVE414                        OFFICER, OPENAI415416    Mr. Altman. Anyway, thank you, Chairman. Thank you, Ranking417Member Cantwell. Thank you, all senators and fellow panelists.418It is a real honor to be here.419    I was here about two years ago and at that time ChatGPT had420recently launched. It was a curiosity in the world. People were421not sure what it was going to mean, what it was going to be422used for.423    Today, we have made significant progress. ChatGPT is used424by more than 500 million people a week. I just saw yesterday425that according to Similarweb it is now the fifth biggest426website on the Internet globally and growing very quickly, but427most of all, it is being used in really important ways.428    It is significantly increasing productivity. We hear429scientists say they are two or three times more productive than430they could be before.431    We hear people that are getting medical advice or learning432in ways they could not before and it is really--it is no longer433this thing that was going to come in the future but it is here434now and people are really using it. We are very proud to be one435of the leaders of this.436    We are very proud that America is leading in AI so437significantly and I think that is critical. What Senator Cruz438said about the importance of innovation in America and that we439have the--what happened with the Internet we have happen again.440    I believe this will be at least as big as the internet,441maybe bigger. That needs to happen. For that to happen442investment in infrastructure is critical.443    I believe the next decade will be about abundant444intelligence and abundant energy, making sure that those--that445America leads in both of those, that we are able to usher in446these dual revolutions that will change the world we live in I447think in incredibly positive ways. It is critical.448    I got to go to Abilene, Texas yesterday where we are449building out what will be the largest AI training facility in450the world. It is coming along beautifully. Super exciting to451see. We need a lot more of that.452    There is a whole sort of AI factory, like, a supply chain453of energy chips, standing up data centers, building the racks454and more.455    We have got to do that really well in the U.S. so that we456can continue to innovate, continue to lead, and continue to,457sort of, shape this revolution.458    Speaking of that, I was very inspired by what Chairman Cruz459said so I would like to deviate from script here and tell a460story. In my prepared written testimony I covered the basics.461    So if it is OK I would love to tell you a story. I grew up462in St. Louis and I was a computer nerd, and it was the time of463the Internet boom and I thought it was the coolest thing ever.464    We kind of lived in this beautiful, old brick house in this465suburb of St. Louis and I lived in the attic, and I had this466computer and I would stay up all night and I would learn to467program, and I got to kind of use the Internet and it was,468like, a crazy time of tons of innovation. All sorts of stuff469was happening.470    It was amazing and it was all happening here. All the471Internet companies were in the U.S. I used a Mac that was built472here. I used chips that were started, you know, near where I473now live.474    And I learned about computers. I thought it was the coolest475thing ever, and I can draw a straight line from that experience476to founding OpenAI and getting to work on companies like477Helion.478    The spirit of American innovation and support of479entrepreneurship. I do not think the Internet could have480happened anywhere else and if that did not happen I do not481think the AI revolution would have happened here.482    I am a child of the Internet revolution. I have the great483honor to be one of the parents of the many parents of the AI484revolution and I think it is no accident that that is happening485in America again and again and again.486    But we need to make sure that we build our systems and that487we set our policy in a way where that continues to happen.488    I think this is magic. I do not want to live in Europe489either. I think America is just an incredible and special490thing, and it will not only be the place where the AI491revolution happens but all the revolutions after.492    I was home visiting St. Louis recently. Drove by our old493house and I kind of, like--it was at night and I looked up and494in that, like, top floor window the light was on and I thought,495you know, hopefully there is some kid in there staying up late496at night playing with ChatGPT, figuring out how he or she is497going to start whatever company comes next and whatever the498next thing is after AI will happen here, too.499    That is, to me, the magic of this country. It is incredibly500personally important and I hope it keeps going.501    Thank you very much for having me.502    [The prepared statement of Mr. Altman follows:]503504             Prepared Statement of Sam Altman, Co-Founder505                  and Chief Executive Officer, OpenAI506    Thank you, Chairman Cruz, Ranking Member Cantwell, and Members of507the Committee.508    I'm Sam Altman, Chief Executive Officer of OpenAI. It is an honor509to return to the Senate and share our view of where AI is today and510where we see it going.511    OpenAI is not a normal company and never will be.512    Our mission is to ensure that artificial general intelligence--513AGI--benefits all of humanity. AGI is a weakly defined term, but514generally speaking we mean it to be a system that can tackle515increasingly complex problems, at human level, in many fields. When we516formed OpenAI more than 10 years ago, we stared at each other around a517kitchen table, wondering how to get started. AI then was a niche tool518for researchers, not the general public.519    In 2016, Chairman Cruz convened his first AI hearing, and my co-520founder, Greg Brockman, testified that AGI models were probably between52110 and 100 years away. Today, the science of AI has advanced so522significantly that we are now confident that we'll reach that milestone523during President Trump's time in office.524    Throughout history, people have crafted tools to scale our525abilities--and we believe AGI will be the most powerful tool ever526created. It will enable people to build incredible things for each527other and improve their quality of life.528    But AGI's full potential won't be realized unless it's safe. The529same capabilities that will enable AGI to support scientific530breakthroughs and accelerate human progress will also create new risk531areas. That's why we red-team relentlessly and lead the industry in532transparency.533    Ultimately, I believe the good will outweigh the bad by orders of534magnitude, and that AGI will help bring us into what I call the535Intelligence Age--an era when everyone's lives can be better than536anyone's life today.537    This future can be almost unimaginably bright, but only if we take538concrete steps to ensure that an American-led version of AI, built on539democratic values like freedom and transparency, prevails over an540authoritarian one.541    The stakes could not be higher--and Congress is right that the542United States must lead the way.543    At OpenAI, we're committed to the path of democratic AI, and we are544humbled that ChatGPT is being used by more than 500 million people each545week to create, discover, and achieve breakthroughs that were once out546of reach.547    America is a nation of innovators, and we want to supercharge548people's ability to use our technology to make their lives better.549    We want to open source very capable models.550    We want to give our users a great deal of freedom in how they use551our tools, and let them personalize ChatGPT to best meet their needs.552    We want to build a brain for the world and make it super easy for553people to use it, with common-sense restrictions to prevent harm.554    And the truth is that AI is already changing the world for the555better.556    Scientists at the U.S. National Laboratories--including Oak Ridge557National Laboratory, Los Alamos National Laboratory, Argonne National558Laboratory, the Princeton Plasma Physics Laboratory, and the Pacific559Northwest National Laboratory--are using our reasoning models to560accelerate breakthroughs in areas like energy.561    In Pennsylvania, ChatGPT is helping state employees do562administrative tasks more quickly, freeing up more time to improve the563delivery of public services.564    And universities in states like Texas, North Carolina, and565California are putting ChatGPT in the hands of students and educators566to build an AI-ready workforce.567    AI will be vitally important to ensuring that today's students are568ready for tomorrow's jobs. In the US, more than one-third of college-569aged young people use our models, mainly for learning and tutoring.570Around the world, most ChatGPT users are under age 35.571    We're proud to offer free access to a technology that is doing so572much for so many people, but AI's biggest gains are still to come.573    Our work at OpenAI suggests that as AI advances, progress574accelerates and becomes increasingly affordable, as reflected in these575three scaling principles:576    Investing more in AI will continue to make it better and more577capable. The intelligence of an AI model roughly equals the log of the578resources used to train and run it. Until recently, scaling progress579has primarily come from training compute and data, but we have shown580how581    to make intelligence scale from inference compute, as well. The582scaling laws that predict these gains are incredibly precise over many583orders of magnitude. It follows that further investment will lead to584further gains, and further benefits to society: We believe that the585socioeconomic value of linearly increasing intelligence is super-586exponential in nature.587    The cost to use a given level of AI capability falls by about 10x588every 12 months, and lower prices lead to much more use. We saw this in589the change in token cost between GPT-4 in early 2023 and GPT-4o in mid-5902024, where the price per token dropped about 150x in that time period.591Moore's Law predicted that the number of transistors on a microchip592would double roughly every two years; the decrease in the cost of using593AI is even more dramatic.594    The amount of time it takes to improve an AI model keeps595decreasing. Put another way, AI models are catching up with human596intelligence at an increasing rate. The typical time it takes for a597computer to beat humans at a given benchmark has fallen from 20 years598after the benchmark was introduced, to five years, and now to one to599two years--and we see no reason why those advances will stop in the600near future.601    So what does that mean practically?602    I believe we'll see many major advances over the next three years,603but here are some examples.604    In 2025, we will release AI-powered tools that can handle605sophisticated software engineering, and AI agents that can handle real-606world tasks like making doctor's appointments and helping to run a607business. These agents will be super assistants who can collaborate608with workers in every industry, doctors in all specialties, and609scientists in every field of research.610    In 2026, AI may unlock a new wave of scientific breakthroughs by611designing experiments to tackle America's toughest challenges in612climate, health, and national security.613    And in 2027, AI-powered robotics could push AI-driven productivity614gains into the physical world, handling routine tasks so people can615spend more time on the work and activities they enjoy.616    As AI systems become more capable, people will want to use them617even more. Meeting that demand requires more chips, training data,618energy, and supercomputers.619    Infrastructure is destiny, and we need a lot more of it.620    Earlier this year I joined President Trump and the CEOs of Oracle621and SoftBank to announce the Stargate Project, a $500 billion dollar622investment in American AI infrastructure.623    Since launching Stargate, governments around the world have asked624about bringing AI infrastructure to their countries and how we can625ensure that democratic AI systems become the global standard.626    In response, we're offering a new kind of partnership--OpenAI for627Countries--to help these countries build up their data center capacity628and ecosystems of AI start-ups and developers. In exchange, these629countries would invest in the Stargate Project--and thus in continued630US-led AI leadership and a global, growing network effect for631democratic AI.632    To close on a personal note, I grew up in St. Louis, part of a633close-knit and competitive family that played 20 Questions to guess634what we were having for dessert. When I was eight, my parents bought me635a Mac LC II. The computer was a literal dividing line in my life. There636was the time before I had a computer, and there has been the time637after. I believe that AI will play a similarly formative role for kids638across the country, including my own.639    I want to thank Chairman Cruz, Ranking Member Cantwell, and the640members of this Committee for your continued leadership on AI. I641appreciate the opportunity to testify today and look forward to642answering your questions.643644    The Chairman. Thank you.645    Dr. Su.646647 STATEMENT OF DR. LISA SU, CHIEF EXECUTIVE OFFICER AND CHAIR,648                  ADVANCED MICRO DEVICES (AMD)649650    Ms. Su. Chairman Cruz, Ranking Member Cantwell, members of651the Committee, it is a real honor to be here on such an652important topic.653    I am Chair and CEO of AMD. We are a U.S. headquartered654semiconductor company founded 56 years ago and we build high-655performance computing chips for the modern economy.656    Every day billions of people rely on our products and657services powered by our technologies but our chips are also658extremely important to support the critical missions including659powering defense systems and secure communications as well as660enabling breakthrough scientific research.661    I have to say our proudest moments, though, are when we see662amazing public-private partnerships and our work in663supercomputing is an example of that.664    Through more than a decade of partnership with the665Department of Energy, AMD now powers the two fastest666supercomputers in the world, one that is housed at Oak Ridge667National Labs that was put into place in 2021 and the other at668Lawrence Livermore National Labs that was just recently put669into commission late last year.670    These systems are really critical from a national671infrastructure standpoint and solve many, many large research672issues as well as national security and scientific leadership.673    Now, in terms of AI, you know, there is so much that has674been stated about AI. I really want to thank Chairman Cruz and675Ranking Member Cantwell for having this hearing. I think it is676a wonderful opportunity to talk about how we win.677    AI is truly the most transformative technology of our time.678The United States leads today, but what I would like to say is679it is a race. Leadership is absolutely not guaranteed. It is a680global race that will shape the outcome of national security681and economic prosperity for many decades to come.682    Now, maintaining our lead actually requires excellence at683every layer of the stack so I am really honored to be here with684my panelists as well.685    We have deep partnerships with Microsoft and OpenAI that686demonstrate how you need silicon, you need software, you need687systems and, really, the application layer to be successful.688    Now, in terms of what to do, I thought about what would be689the most important things to say today and I put them in five690categories.691    I think the first and probably the foremost is we must692continue to run faster. This is a race and the race does not693stand still. Nobody in the world stands still.694    We lead today because of the bold decisions that we have695made and because of the innovation economy that we have. But we696need to continue to run faster and that means ensuring that we697have computing available.698    I think Sam's story about Abilene is an excellent example699of how when you allow the computing infrastructure to expand at700the rate and pace that the private sector wants, you actually701make tremendous progress.702    I would also like to mention the importance of open703ecosystems. I think open ecosystems are really a cornerstone of704U.S. leadership and that allows, you know, frankly, ideas to705come from everywhere and every part of the innovation cycle,706reducing barriers to entry and strengthening security as well707as creating, frankly, a competitive marketplace for ideas.708    Third, we are very happy to see the focus on a robust709domestic supply chain. For us in the semiconductor world we710used to not get so much attention.711    Now we get a lot of attention thanks to the importance of712chips, and the fact is we need more manufacturing in the U.S.713The efforts so far have made good progress but there is a lot714more that can be done and that should be done in public-private715partnership.716    Fourth, we must invest in talent. Frankly, the United717States should be the best place to study AI, to work in AI, to718really move forward all of the innovations that we need and I719think, again, this can also be done in significant public-720private partnership.721    And then fifth, of course, in the area of export controls722we must--we totally understand as an industry the importance of723national security and that is, you know, without--that goes724without saying as a U.S. company.725    But we also want to ensure--as Chairman Cruz and Ranking726Member Cantwell stated, it is important to have widespread727adoption of U.S. technologies. We lead today because we have728the best technology.729    However, if we are not able to fully have our technology730adopted in the rest of the world there will be other731technologies that will come to play. They may not be as good as732we are today but, frankly, usage really spurs innovation and733this is something that we certainly need to work with in734public-private partnership.735    And I would, frankly, end by saying, you know, like Sam I736had a computer when I was growing up. I grew up in New York. I737am a little older than Sam so my first computer was a Commodore73864 and then I graduated to the Apple II. But the fact is this739is the best place to do computing innovation in the world. We740want it to stay that way with, really, a very rich and broad741ecosystem.742    So thank you again for the opportunity to be here today.743    [The prepared statement of Ms. Su follows:]744745           Prepared Statement of Dr. Lisa Su, Chair and CEO,746                   Advanced Micro Devices Inc. (AMD)747    Chairman Cruz, Ranking Member Cantwell, Members of the Committee,748thank you for the opportunity to speak with you at such a consequential749moment.750    I am Chair and CEO of AMD, a U.S.-headquartered semiconductor751company founded 56 years ago. We build high-performance computing chips752that power the modern economy.753    Every day, billions of people rely on products and services powered754by AMD technologies. AMD chips also play a vital role advancing many of755our Nation's most critical missions, from powering defense systems and756secure communications to enabling breakthrough scientific research,757medical innovations, and quantum computing.758    Our work in supercomputing showcases the full strength of AMD's759innovation and public impact. Through more than a decade of partnership760with the Department of Energy, AMD now powers the world's two fastest761supercomputers: Frontier, which went into operation at Oak Ridge762National Labs in 2021, and El Capitan, which went into operation at763Lawrence Livermore National Labs late last year. These systems are764critical infrastructure for U.S. national security and scientific765leadership, including the latest advances in drug discovery, medical766research, climate research, hypersonic flight, and even training future767generations of more capable AI models.768    Today we are here to talk about AI. No technology today better769demonstrates the power of high-performance computing than AI.770    AI is the most transformative technology of our time. The United771States leads today, but leadership is not guaranteed. This is a global772race, and the outcome will shape economic growth, national security,773and technological influence for decades to come.774    Maintaining our lead requires excellence at every layer of the AI775stack. AMD's collaborations with Microsoft and OpenAI demonstrate how776industry leaders can work together across hardware, software, and777systems to advance state-of-the-art AI.778    Underneath every model, every breakthrough, and every application779is massive amounts of computing power. If we want to lead in AI, we780must lead in the infrastructure that powers it. That requires urgency781across five national priorities.782    First, we must keep running faster. America leads when it moves783fast and thinks big. From semiconductors to the internet, speed has784turned bold American ideas into global industries. In AI, speed785requires accelerating chip and system innovations that deliver more786performance with greater efficiency. It also means making AI compute787infrastructure readily available across the industry. This will require788rapidly building data centers at scale and powering them with reliable,789affordable, and clean energy sources. Moving faster also means moving790AI beyond the cloud. To ensure every American benefits, AI must be791built into the devices we use every day and made as accessible and792dependable as electricity. From vehicles and sensors to PCs and medical793tools, bringing the power of AI to every enterprise and every American794will enable faster decisions, smarter systems, and better services795where they matter most. We have the technology, intellectual property,796and talent to do that today, but it is a global race and we must keep797accelerating our pace.798    Second, we must champion open ecosystems. Open standards have long799been a cornerstone of U.S. leadership. The same approach must guide our800path with AI as well. Open ecosystems allow hardware, software, and801models from different vendors to work together. This accelerates802innovation, reduces barriers to entry, strengthens security through803transparency, and creates healthier, more competitive markets.804    Third, we must build a robust domestic supply chain for advanced805semiconductor manufacturing and packaging. AI leadership depends on the806ability to build complete, integrated systems. That means ensuring we807have domestic capabilities in both wafer manufacturing at the most808advanced nodes and next-generation packaging technologies as well as809the advanced system capabilities needed to bring it all together. AMD810is proud to be one of the first partners producing leading-edge chips811at TSMC's new fab in Arizona. The domestic semiconductor manufacturing812projects announced to date represent meaningful progress, but there is813much more that we can do. This is an area where strong public-private814partnerships are critical. The entire semiconductor industry is aligned815on the need to work together and partner with the government to816significantly scale U.S. chip production and advanced packaging817capabilities here at home.818    Fourth, we must invest in talent and ensure our national strategy819for STEM education, workforce training, and immigration supports820sustained AI leadership. The private sector can certainly do more,821including expanding university partnerships, investing in reskilling822programs, and developing the cross-disciplinary talent required for823success. We should incentivize companies to increase their most824critical AI R&D efforts here at home and ensure our immigration825policies attract and retain the world's best AI talent. We should make826America the absolute best place for AI talent in the world.827    Fifth, we must balance the need for national security with the828imperative to enable the widespread adoption of U.S. technologies. As829the government considers policies like AI diffusion, it is important to830remember that the U.S. leads in AI today and we want the rest of the831world building on our platforms. If our international partners cannot832access U.S. platforms, they will adopt alternatives that may be less833advanced today but will mature over time. Threading this needle834requires closer collaboration between government and industry to ensure835rules are clear, consistent, and aligned with both competitiveness and836security.837    This is a pivotal moment. A once-in-a-generation opportunity to838secure U.S. leadership in AI and advanced computing. This is not just839about developing a transformative technology. It's about shaping the840future of our economy, safeguarding our national security, and841enhancing our global competitiveness.842    Now is the time to ensure the United States doesn't just keep up,843but takes the decisive steps needed to cement our leadership.844    Thank you again. I look forward to your questions.845846    The Chairman. And I had an Apple II as well with a shoe box847of floppy disks and somehow I ended up taking a wrong turn and848ending up in politics instead.849    Mr. Intrator.850851 STATEMENT OF MICHAEL INTRATOR, CO-FOUNDER AND CHIEF EXECUTIVE852                       OFFICER, COREWEAVE853854    Mr. Intrator. I started out with a VIC-20.855    Chairman Cruz, Ranking Member Cantwell, and distinguished856members of the Committee, thank you for the opportunity to857testify today. I am honored to appear alongside my industry858colleagues and partners.859    My name is Michael Intrator. I am the Co-Founder and CEO of860CoreWeave founded 7 years ago. CoreWeave started like many861innovative ventures, humbly in a garage, experimenting,862initially with graphics processing units, or GPUs, for863cryptocurrency mining.864    Recognizing the transformational potential, we pivoted to865support powerful AI applications, dramatically scaling the866vision and operation.867    Today, CoreWeave stands at the forefront of America's AI868infrastructure revolution, operating more than 30 data centers869across 15 states. We manage more than 250,000 GPUs currently870using 360 megawatts of power.871    Over two short years our revenue has surged by 12,000872percent, reaching $1.9 billion in 2024. As a result of this873progress, CoreWeave became a publicly traded company on March87428th of 2025.875    CoreWeave's rapid growth is a testimony not only to the876technology but also to the surging global demand for advanced877AI infrastructure. Our infrastructure enables American878businesses to rapidly translate AI aspirations into impactful879economic realities.880    By empowering companies to accelerate innovation we are881fueling America's competitive edge while improving productivity882and prosperity.883    Modern AI requires specialized infrastructure, purpose884built computing capabilities that surpass traditional cloud885computing in scale and performance, today's general purpose886cloud that was built to support and scale the complexity of AI887workloads.888    We cannot run a 21st century economy on the 20th century's889infrastructure. AI workloads involve trillions of simultaneous890calculations demanding unprecedented computing power, advanced891cooling systems, cutting-edge chip technology, ultra high-speed892networks and accelerated storage.893    Since 2018 the computing power necessary for advanced AI894models has multiplied approximately 100,000 fold. At CoreWeave895our facilities symbolize America's great tradition of896innovation.897    Our data centers built, maintained, and staffed by skilled898American workers embody how modern technology not only899stimulates economic growth and enhances national security, but900also improves humans' lives.901    We are at a critical juncture in the global AI competition.902The nation that leads in infrastructure will set the global903economic agenda and shape human outcomes for decades.904    Our largest competitor, China, recognizes the stakes and is905spending significant resources to strengthen their position. I906want to focus on four elements of policy that will help907determine whether the U.S. secures its leadership role in the908AI race.909    First, strategic investment stability. AI infrastructure is910deeply capital intensive and requires a significant level of911coordination across industry stakeholders.912    Stable predictable policy frameworks, secure supply chains,913and regulatory environments that foster innovation are crucial.914Policy makers must provide clear and consistent policy and915regulations across all jurisdictions that enables long-term916investment and rapid scaling of AI technology.917    Second, energy infrastructure development. To support the918rapid deployment of AI infrastructure America must ensure919abundant and affordable supplies of energy. Careful reforms and920permitting and regulatory process are necessary to accelerate921infrastructure projects, facilitate more rapid construction,922interconnections, and energy for data centers.923    Third is global market access. Maintaining America's924leadership also means ensuring our technology has fair access925to global markets.926    Export controls and trade agreements can be calibrated to927both address national security risks and support global928diffusion of American AI technology.929    And, finally, public-private partnerships and workforce930development. America's unique advantage in the AI race is931enhanced by our powerful tradition of public-private932partnership.933    CoreWeave is proud to co-found the New Jersey AI hub with934Microsoft, Princeton University, and the New Jersey economic935development initiative.936    Initiatives like this develop critical workforce skills,937foster innovation, and ensure economies and communities are938prepared for the AI-driven future.939    America stands ready to lead the AI revolution, which will940bring enormous benefits. It is a rare moment in time that we941must meet.942    If government, industry, and all affected parties work943together the United States can win this race and seize the vast944opportunity ahead of us.945    Thank you again for the opportunity to testify. I look946forward to answering your questions.947    [The prepared statement of Mr. Intrator follows:]948949          Prepared Statement of Michael Intrator, Co-Founder950                 and Chief Executive Officer, CoreWeave951Introduction952    Chairman Cruz, Ranking Member Cantwell, and distinguished Members953of the Committee, I want to thank you for the opportunity to provide954testimony today on how we can ensure the United States remains the955global leader in Artificial Intelligence (AI) innovation.956    Today we're on the verge of the next great revolution of AI: a957technology dramatically reshaping industries, innovation, and958productivity at a massive scale through systems of unprecedented959complexity. Millions of hours of training, billions of inference960queries, trillions of model parameters, and continuous dynamic scaling961are all driving an insatiable hunger for compute and energy that962borders on exponential.963    At CoreWeave, we are not only deploying the infrastructure that964hosts some of the most massive and powerful AI models in existence, but965also doing it at scale.966    I appreciate the opportunity to describe CoreWeave's journey and967our role in this historic race.968CoreWeave's Role969    CoreWeave powers AI innovations by bridging the gap between AI970ambition and execution. CoreWeave provides the cloud platform, purpose-971built for AI, that delivers the speed, performance, and expertise972needed to unleash AI's full potential.973    To train their next generation of AI models, AI labs require974compute resources. And to maintain their leadership position at the975forefront of AI innovation, researchers demand the latest and most-976performant computing infrastructure. They leverage these compute977resources to run the trillions of operations that operate algorithms978and process data to train the next generation of models.979    The ``Scaling Law'' \1\ has demonstrated that increasing the980compute deployed against models translates to better performance. The981relationship is exponential. Orders of982---------------------------------------------------------------------------983    \1\ Scaling laws describe how the performance of AI systems improve984as the size of the training data, model parameters, or computational985resources increases.986---------------------------------------------------------------------------987    magnitude of increased compute are required to unlock incremental988gains in model performance. For example, in 2018, one of the most989popular and advanced generative AI models required a certain amount of990compute to train. Just seven years later, in 2025, the amount of991compute needed for the latest frontier models grew by about 100,000992times--an extraordinary increase.\2\ Performance, usefulness, and real-993world adoption has increased dramatically. This demonstrates that the994limiting boundary of AI development is high-performance compute,995delivered at scale and operated at peak efficiency.996---------------------------------------------------------------------------997    \2\ Epoch AI, ``Training Compute of Frontier AI Models Grows by 4-9985x per Year,'' May 28, 2024.999---------------------------------------------------------------------------1000    To access the increased need for compute and to avoid being left1001behind, enterprises and labs are continuously demanding deployments of1002the latest generation chips at larger scales. In just the two short1003years between the most recent generations of Nvidia chips, training1004performance increased by 4x from Hopper to Grace Blackwell.\3\1005---------------------------------------------------------------------------1006    \3\ Nvidia, ``NVIDIA GB200 NVL72: Powering the new era of1007computing.''1008---------------------------------------------------------------------------1009    CoreWeave enables AI labs, platforms, and enterprises to move the1010boundaries of compute forward. Our end-to-end cloud platform is1011purpose-built for the scale, performance, and expertise needed to power1012AI innovation and meet the demands of accelerated computing. We1013construct and power data centers, and provision high-performance1014computing infrastructure which enables enterprise and labs to access1015these essential resources. Our expertise is in managing a complex and1016fragile ecosystem across supply chain, energy, financing, and1017technology partnerships to build, optimize, and deploy our platform at1018scale.1019    As of December 2024, we operated in 32 data centers in the U.S. and1020Europe, deploying more than 250,000 GPUs utilizing 360MW of active1021power. We have now contracted for 1.3GW of power. Adequate, reliable1022supplies of power are essential to drive this revolution and for the1023U.S. to win this race.1024    This year, CoreWeave was the sole cloud to be ranked Platinum and1025as the #1 leader in AI cloud performance, as attributed by1026SemiAnalysis's Platinum ClusterMAXTM Rating.\4\ And we have1027established a track record as among the first to market with the latest1028generations of hardware, such as Nvidia's most recent GB200 NVL72 chip,1029which leading labs, such as IBM, Mistral AI, and Cohere, are already1030using to improve and accelerate their training jobs.\5\ We also support1031OpenAI--our strategic deal of nearly $12B provides compute capacity for1032training and delivering its latest models at scale to its hundreds of1033millions of users around the world.\6\1034---------------------------------------------------------------------------1035    \4\ CoreWeave, ``CoreWeave Ranks as #1 AI Cloud, Backed by1036SemiAnalysis' Platinum ClusterMAXTM Rating,'' April 10,10372025.1038    \5\ Nvidia, ``Thousands of NVIDIA Grace Blackwell GPUs Now Live at1039CoreWeave, Propelling Development for AI Pioneers,'' April 15, 2025.1040    \6\ CoreWeave, ``CoreWeave Announces Agreement with OpenAI to1041Deliver AI Infrastructure,'' March 10, 2025.1042---------------------------------------------------------------------------1043    CoreWeave is purpose-built for the demands of accelerated1044computing. We deliver this infrastructure with cutting-edge performance1045and scale, and provide the expertise with the infrastructure that AI1046needs today and in the future. And as a result, our customers are able1047to train earlier, build quicker, and get to market faster, which is1048critical for the U.S. to maintain its lead in AI. CoreWeave is the1049engine that will propel the U.S. forward in the AI race.1050CoreWeave's History1051    CoreWeave began as many start-ups and great entrepreneurial1052companies do--in a garage with an idea, which was to try leveraging1053GPUs for crypto mining. CoreWeave's founders purchased their first GPU1054in 2016, which turned into hundreds, then thousands, and now hundreds1055of thousands. Over the course of the next few years, CoreWeave began1056looking for opportunities to use its fleet of GPUs for other high-1057performing use cases beyond crypto mining, such as visual effects1058(VFX), and then AI.10591060   In 2020, CoreWeave launched as the ``world's first1061        specialized cloud.''10621063   In 2021, CoreWeave operated the largest Nvidia A40 fleet in1064        North America.10651066   By 2022, the world began to realize that more compute was1067        required to scale AI model training. CoreWeave began to scale1068        even more rapidly.10691070   By 2023, the company had three data centers running more1071        than 17,000 GPUs with approximately 10 megawatts (MWs) of1072        active power.10731074   By 2024, CoreWeave had ten data centers running more than1075        53,000 GPUs with more than 70MW of active power.10761077   And, one year later, at the end of 2024, we had 32 data1078        centers running more than 250,000 GPUs with approximately 360MW1079        of active power.10801081    CoreWeave became a publicly traded company on March 28, 2025.1082    This progress occurred in five short years. And that is the speed1083which is required to drive this technological revolution. Most1084recently, CoreWeave completed its acquisition of Weights & Biases, a1085leading AI developer platform. Our vision is that CoreWeave + Weights &1086Biases will deliver the leading AI Cloud Platform--purpose-built to1087develop, deploy, and iterate AI faster. Together, we will enable1088faster, more efficient AI development, empower AI developers to1089innovate quickly, provide seamless integrations for AI development, and1090support the world's most advanced AI innovators to unleash AI's full1091potential.1092The Need for AI Infrastructure and Re-platforming1093    AI requires a fundamentally different computing infrastructure from1094the existing one. Training state-of-the-art models and running1095inference at scale requires trillions of simultaneous calculations1096across billions of parameters. To fulfill this requirement, high-1097performing compute infrastructure necessitates a more concentrated1098power footprint, increased cooling needs, the latest chips, high-1099throughput networking, accelerated storage, and more.1100    In contrast, the generalized clouds that serve the world today were1101not built to serve the specific requirements of AI. These cloud1102platforms were built for day-to-day web hosting, database management,1103and running SaaS applications--workloads that rely on simple, fixed-1104logic calculations and lightweight processing.1105    Operating compute at scale and at the intensity of AI is highly1106complex. There are significant inefficiencies associated with operating1107AI workloads ranging from hardware failures to scheduling optimal1108usage. A single 32,000 GPU cluster may require the deployment of1109approximately 600 miles of fiber cables and approximately 80,000 fiber1110connections, along with highly-specialized heat management capabilities1111to support high-power density. Each of these variables increases the1112number and complexity of possible failure points. When a cluster1113suffers a component failure (GPU, network, memory, cable, cooling,1114etc), it can adversely impact the entire cluster by reducing training1115performance, or even causing the entire project to fail.1116    The difficulty in managing large clusters leads to what we call the1117``AI Efficiency gap,'' which we evaluate based on Model FLOPs1118Utilization (MFU). This is a measure of the observed throughput1119compared to system maximum if the system were operating at peak1120capacity. Typically, the complexity of managing AI infrastructure means1121that a majority of the compute capacity in GPUs can be lost to system1122inefficiencies, with empirical evidence suggesting observed levels of1123performance in the 35 percent to 45 percent range.1124    As a result, the world is undergoing a ``re-platforming'' from1125traditional generalized cloud computing infrastructure to AI cloud1126computing infrastructure. And to achieve this, cloud platforms are1127being fundamentally reimagined, with every layer of the technology1128stack being specifically optimized for AI workloads. This is the1129purpose-built computing infrastructure needed to support the scale and1130complexity of AI workloads.1131CoreWeave's Cloud Platform1132    We have built our platform for the new requirements of AI cloud1133computing infrastructure.1134    CoreWeave's cloud platform is an integrated solution that is1135purpose-built for running AI workloads such as model training and1136inference at superior performance and efficiency. It includes1137infrastructure services, managed software services, and application1138software services, all of which are augmented by our proprietary1139Mission Control and observability software. This proprietary software1140enables the provisioning of infrastructure, the orchestration of1141workloads, and the monitoring of our customers' training and inference1142environments to ensure high availability and minimize downtime.1143    To unlock the full potential of AI infrastructure, CoreWeave helps1144to bridge the MFU ``efficiency gap'' between the observed 35-45 percent1145and the theoretical 100 percent, driving as much as 20 percent higher1146performance than public benchmarks.\7\ To achieve this, performance1147optimizations are built into every layer of the platform to enhance1148distributed training throughput. And our ability to close this gap1149significantly enhances performance, improves model quality, accelerates1150development timelines, and reduces overall AI model costs.1151---------------------------------------------------------------------------1152    \7\ CoreWeave, ``CoreWeave leads the Charge in AI Infrastructure1153Efficiency, with up to 20 percent Higher GPU Cluster Performance than1154Alternative Solutions,'' March 19, 2025.1155---------------------------------------------------------------------------1156    What does this mean for the U.S.? As we improve our efficiency, and1157close this gap, the United States will maintain its edge in the global1158AI race, stimulating economic activity, and enhancing national security1159while improving the provision of essential services for all. This is1160what the race is all about.1161U.S. Global Leadership in AI1162    We stand at a critical inflection point in the global AI race,1163representing a pivotal moment that will influence economic prosperity,1164national security, technological standards and how we provide essential1165services to all Americans. AI represents the next major evolution of1166technology with the potential to transform society. This is America's1167AI moment, and a strategic opportunity America cannot afford to miss.1168    Economic Prosperity: AI is projected to generate a cumulative1169global economic impact of $20 trillion, representing 3.5 percent of1170global GDP, by 2030.\8\ The country that leads this transformation will1171capture a disproportionate share of this new economic frontier. If1172America maintains global leadership in AI, the productivity gains, new1173products, high-value jobs, and breakthroughs across industries from1174healthcare to manufacturing created by AI will help drive prosperity1175across the American economy benefitting all people.1176---------------------------------------------------------------------------1177    \8\ IDC, ``The Global Impact of Artificial Intelligence on the1178Economy and Jobs,'' September 2024.1179---------------------------------------------------------------------------1180    National Security: As advanced AI capabilities become essential to1181modern defense including improvements to weapons systems and1182battlefield capabilities, intelligence, and cybersecurity systems,1183maintaining America's technological edge becomes inseparable from our1184national security. Falling behind is not an option when other countries1185are rapidly advancing their own AI capabilities with explicit aims to1186challenge American global economic and military leadership.1187    Shaping the Future of AI: The country that leads AI development1188will shape how this technology evolves globally. The standards,1189protocols, and ethical frameworks that will govern AI will reflect the1190values of whichever country wins this race.1191    The foundational AI infrastructure being built today will help1192determine where AI development occurs. Success in the global AI race1193will increasingly depend on purpose-built AI computing infrastructure,1194not just general-purpose systems deployed at scale. Nations that1195successfully ``re-platform'' gain compounding advantages in model1196capabilities and development speed.1197    CoreWeave is at the forefront of developing the purpose-built1198infrastructure that powers America's AI capabilities. Leading companies1199and AI labs such as IBM, Mistral, and Cohere rely on CoreWeave's1200infrastructure. Our success supports broader national objectives by1201ensuring the U.S. maintains the world's most advanced computing1202infrastructure which is required to drive AI.1203Factors Critical to Continued U.S. Leadership in AI Infrastructure1204    America's leadership position in AI depends in part on maintaining1205its edge in the underlying infrastructure that drives it. Based on1206CoreWeave's experience building and operating AI computing1207infrastructure, I would like to highlight several critical areas that1208will determine if our Nation maintains its leadership position. Many of1209these areas focus on the critical elements of policy which will impact1210how this sector evolves.1211Strategic Investment Stability1212    AI infrastructure investment requires a significant level of1213coordination across multiple industry and government stakeholders due1214to the scale and timeline of these projects, representing substantial1215capital commitments with years-long development and operational1216horizons. CoreWeave benefits from robust collaborations with leading1217chipmakers, original equipment manufacturers (``OEMs''), and software1218providers to supply us with infrastructure components and other1219products. The highly specialized infrastructure that is required to1220unlock the potential of AI is immensely challenging to build and1221operate, especially at scale. This requires: (i) tens of thousands of1222GPUs, (ii) thousands of miles of high-speed networking cables, (iii)1223hundreds of thousands of interconnects coming together to create1224``superclusters'' for training and serving AI models, and (iv) hundreds1225of MWs of power and substantial amounts of storage.1226    To sustain U.S. leadership in AI, it is important for U.S. AI cloud1227computing companies to maintain access to a reliable supply chain1228necessary to access all of the components necessary to develop and run1229the cutting-edge AI infrastructure. Acquiring these necessary high-1230performance components to power AI workloads requires managing a1231complex global supply chain and maintaining robust supply chain1232relationships. Continued engagement with leading global suppliers and1233strategic partners is vital to ensuring the continued operation,1234expansion, and rapid deployment of U.S. AI infrastructure and to uphold1235U.S. competitiveness. Predictable policy is essential for this.1236    Significant private sector investment and development has helped1237the United States establish an early and important lead in AI1238infrastructure. The U.S. accounts for roughly 40 percent of the global1239market for data center capacity, with six of the top ten markets.\9\1240---------------------------------------------------------------------------1241    \9\ Cushman & Wakefield Research, ``2024 Global Data Center Market1242Comparison.''1243---------------------------------------------------------------------------1244    The importance of AI and U.S. leadership is not lost on our1245competitors. Intensifying global competition for AI infrastructure1246demands that this initial lead must be carefully and actively1247maintained. The European Union launched its AI Continent Action Plan in1248April setting out ambitious goals to triple data center capacity across1249member states in the next five to seven years. The EU also announced a1250=20 billion investment into five gigafactories-massive high performance1251computing facilities equipped with approximately 100,000 state-of-the-1252art AI chips--and reforms related to permitting, energy issues and1253water usage.\10\ China has made its ambitions regarding AI clear1254through coordinated national strategies and streamlined deployment1255timelines that can compress years into months in their effort to shrink1256America's current AI lead.1257---------------------------------------------------------------------------1258    \10\ European Commission, ``The AI Continent Action Plan,'' April12599, 2025, https://digital-strategy.ec.europa.eu/en/library/ai-continent-1260action-plan.1261---------------------------------------------------------------------------1262    Countries around the world are aggressively pursuing coordinated AI1263strategies, implementing policies which subsidize infrastructure, and1264accelerate deployment timelines. In this high-stakes environment, the1265capacity of American companies to build AI infrastructure swiftly and1266with assurance will be a decisive factor in the AI race and determine1267whether the United States retains its leadership position.1268    Sustained American leadership in AI infrastructure faces potential1269headwinds from multiple sources of uncertainty. These include1270volatility in the global supply chain for critical components, such as1271advanced semiconductors and networking equipment which can disrupt1272deployment timelines. These fluctuations can lead to delays or1273unanticipated cost overruns, adversely affecting American companies'1274ability to rapidly scale AI capabilities.1275    Changes in regulation at both the Federal and state levels1276introduce substantial uncertainty for leading businesses making1277investment decisions. Changes in export controls, energy policies, the1278potential need to add gigawatts of power to meet increased demand, and1279the emerging landscape of AI-specific regulations at different levels1280of government also affect the pace and scope of infrastructure1281deployment. The lack of regulatory clarity can deter investment and1282slow down the innovation cycle. Additionally, American companies will1283be affected by the rules and institutions being developed around the1284world both in individual nations and in important forums in which key1285competitors participate that will govern the use of AI. As the world's1286dominant economic player and technological leader, it is important for1287the U.S. government to drive the rules which shape the future playing1288field for American companies.1289    Finally, the potential for a fragmented regulatory framework, with1290differing requirements from state to state and potentially at the1291Federal level, poses a unique challenge. For instance, inconsistent1292definitions of key terms which define various AI activities across1293jurisdictions could force companies to navigate a complex web of1294compliance regimes for fundamentally similar activities. These types of1295policies would require participants in the AI infrastructure to1296consider designing alternative products and strategies to do business1297in different jurisdictions. This regulatory patchwork would lead to1298increased costs, operational inefficiencies, and ultimately, a1299competitive disadvantage for American companies in the global AI race.1300These uncertainties disproportionately affect newer entrants like1301CoreWeave and other specialized providers, potentially stifling the1302very innovation that drives American leadership in this critical1303sector.1304    To ensure the United States remains at the forefront of AI,1305American companies must lead AI infrastructure development. This1306requires a coordinated policy strategy to mitigate key uncertainties,1307many of which this section touches upon, maintain appropriate1308oversight, and create a stable, predictable policy environment that1309fosters investment, continued growth and innovation.1310Energy and Infrastructure Development: Powering AI Leadership1311    The race to build AI infrastructure is fundamentally tied to the1312Nation's ability to continue to develop a new generation of data1313centers that drive innovation, to bring them online and ensure there is1314sufficient electricity to power them. This will be affected by the1315processes and permitting systems that are used to develop data centers1316and to develop adequate, reliable supplies of new power generation and1317the interconnection and transmission systems capable of delivering it1318at pace and scale.1319    The dual challenges of adding new power supplies and streamlining1320infrastructure development, are not merely logistical hurdles, but1321critical factors that will determine whether America can maintain its1322global AI leadership. Failure to address these challenges effectively1323risks ceding ground to international competitors, particularly China,1324who are aggressively pursuing their own AI ambitions.1325    Energy considerations are critical to the development and operation1326of AI infrastructure. After a prolonged period of relatively flat1327electricity consumption, according to analysis, the U.S. is now1328experiencing a significant and accelerating increase in power demand.1329This surge is driven by several concurrent trends, including the1330onshoring of new manufacturing facilities, the widespread1331electrification of transportation and heating, and the growth in data1332centers.1333    AI computation is energy-intensive. Training large language models,1334running complex simulations, and deploying AI applications all require1335significant amounts of power. Widespread AI adoption will further1336increase this demand, even as companies continue to innovate and1337improve efficiency. According to a report released by the U.S.1338Department of Energy, data centers consumed approximately 4.4 percent1339of total U.S. electricity in 2023. This figure is projected to rise in1340the coming years, potentially consuming between 6.7 percent and 121341percent of total U.S. electricity by 2028.\11\ This projected increase1342underscores the urgent need for policymakers at all levels of1343government to put policies in place that will enable the development of1344new power supplies and the infrastructure to deliver it. Given that1345these projects can cost in the hundreds of millions of dollars and1346years to implement, there is no time to lose in getting started.1347---------------------------------------------------------------------------1348    \11\ ``2024 United States Data Center Energy Usage Report,''1349Lawrence Berkeley National Laboratory, December 2024.1350---------------------------------------------------------------------------1351    The implications for global AI leadership are clear and1352consequential. Regions that can provide abundant, reliable, and cost-1353effective energy will attract billions of dollars of AI infrastructure1354investment. Conversely, energy constraints, whether in the form of1355limited supply, unreliable delivery, policy uncertainty, or prohibitive1356costs, can and will push development and the associated investment1357elsewhere.1358    CoreWeave's site selection consideration for data centers1359illustrates these priorities:13601361   Availability of abundant, reliable power, and where1362        available, non-emitting sources13631364   Competitive rates13651366   Diverse energy sources13671368   Pathways for capacity expansion13691370   Efficient permitting processes that provide timeline1371        certainty13721373    However, obtaining the necessary approvals to build both energy1374infrastructure and data centers is often a critical bottleneck. Every1375month of delay represents lost ground in a field where the pace of1376innovation is measured in weeks, not years.1377    In particular, it is challenging to develop energy projects.1378Securing the necessary approvals and permits can take a significant1379amount of time. Inefficiencies in the permitting process can1380significantly impact both energy availability and whether attractive1381sites for data center development can move forward. The variability in1382permitting timelines across jurisdictions and the potential for1383multiple, sequential review processes and litigation can increase the1384time required to develop a project leading to delay or potentially1385stopping projects.1386    There will be challenges in streamlining the permitting and1387regulatory processes required to develop the energy and data center1388infrastructure necessary for the U.S. to maintain its leadership in AI.1389Goals in streamlining these processes include:13901391   Maintaining and growing a balanced portfolio of generation1392        powered by diverse energy sources that can meet increasing1393        demand to ensure availability and reliability at reasonable1394        costs13951396   Expanding and modernizing the Nation's transmission systems13971398   Providing developers of data center capacity and associated1399        infrastructure with predictable timelines and reduced wait1400        times for feasibility studies, interconnections, and builds14011402   Streamlining permitting processes while maintaining1403        appropriate oversight14041405    CoreWeave understands that the processes put in place to achieve1406these important objectives need to consider the views of key players1407that will make these investments and the communities in which these1408facilities are located.1409    We hope efforts to streamline the permitting process to enable the1410addition of new sources of generation and the transmission1411infrastructure to deliver it will receive attention by this1412Congress.\12\1413---------------------------------------------------------------------------1414    \12\ CoreWeave is a member of the Data Center Coalition. DCC's AI1415Action Plan submission includes additional discussion related to1416permitting and energy infrastructure, available here. [https://1417static1.squarespace.com/static/63a4849eab1c756a1d3e97b1/t/141867d84a70db36cf08e2a3141929cb/1742228107114/DCC+Comments+-+RFI+AI+Action+Plan.pdf]1420---------------------------------------------------------------------------1421    We recognize that this issue will not be resolved solely at the1422Federal level. All levels of government have a role to play in1423addressing the challenges in adding the necessary infrastructure to1424meet energy requirements. A coordinated effort amongst federal, state,1425and local government, industry, and other affected parties is required1426to address these interrelated challenges which include creating1427efficient, transparent processes that allow infrastructure to be built1428at the pace and scale that technological advancement requires and for1429the U.S. to maintain its dominant position in AI.1430Global Diffusion of the American AI Stack1431    Global market access is a pivotal factor in determining which1432nation will lead in the AI domain. Export controls and trade agreements1433can be designed to achieve multiple objectives: they can facilitate1434legitimate market access for American businesses while also mitigating1435potential national security risks. However, controls that are not1436calibrated can inadvertently bolster foreign competitors by1437incentivizing AI development and deployment outside of the U.S., and1438competitors filling the void left by U.S. firms. This could result in1439the loss of technological expertise and economic benefits.1440    To bolster American AI leadership, export controls and1441international agreements should consider:14421443   Precision Targeting of National Security Risks: Controls1444        should be focused on technologies, entities and nations that1445        pose genuine and demonstrable threats to national security,1446        with clear and specific parameters.14471448   Supporting American Technological Leadership: Restrictions1449        imposed on U.S. technologies and where they can be exported1450        should consider negative impacts on the ability of U.S.1451        companies to compete in global markets. This includes1452        considering the potential for retaliatory measures from other1453        nations and the risk of creating a 'chilling effect' on1454        investment and innovation.14551456   Strategic Alignment with Allies: Close coordination with1457        like-minded international partners is essential to ensure the1458        effectiveness of export controls and prevent the fragmentation1459        of the global AI market. Aligning with allies can also foster1460        the expansion of a collaborative, secure, and trustworthy AI1461        ecosystem.14621463    These considerations are crucial in shaping the future landscape of1464AI innovation. A well-calibrated approach will ensure that the next1465generation of AI development is anchored in the United States,1466leveraging American infrastructure and expertise.1467    Conversely, controls could inadvertently have the effect of1468limiting opportunities for the export of U.S. technology and expertise,1469with adverse impacts. A strategy that carefully differentiates between1470markets, tailors export restrictions to mitigating specific risks, and1471fosters international cooperation, can effectively protect national1472security while simultaneously enhancing America's ability to lead and1473shape the global AI diffusion race.1474Public-Private Partnerships Accelerating American Innovation1475    A unique American advantage in the AI race is our ability to forge1476effective partnerships between government, industry, academia and1477elements of civil society. These collaborations combine the agility,1478ingenuity, expertise and resources of the private sector with the long-1479term vision of the public sector and the basic and applied research1480capabilities of academic and research institutions. This approach helps1481foster an innovation ecosystem that is difficult for competitors to1482replicate.1483    CoreWeave is proud to be a founding partner of the New Jersey AI1484Hub, along with Microsoft, Princeton University, and the New Jersey1485Economic Development Authority. This AI Hub will focus on research and1486development efforts, applications of AI in several industry sectors,1487and AI workforce development and education.\13\1488---------------------------------------------------------------------------1489    \13\ Governor Phil Murphy, ``Governor Murphy, Princeton University,1490Microsoft & CoreWeave Cut Ribbon on Major Artificial Intelligence1491Hub,'' March 27, 2025, https://www.nj.gov/governor/news/news/562025/1492approved/20250327a.shtml.1493---------------------------------------------------------------------------1494    CoreWeave is deeply committed to supporting AI education and1495research, and public-private collaborative partnerships. Similar1496partnerships across the country could further accelerate America's AI1497capabilities, and we encourage policymakers to explore this model of1498collaboration. These types of partnerships also accelerate AI and data1499center workforce development in the US. Worker shortages in the data1500center industry are becoming commonplace as the skills gap widens. Data1501center employers have struggled to find enough trained workers. Half or150250 percent of surveyed data center managers in 2020 reported having1503difficulty finding skilled workers to fill positions, and 71 percent1504continued to report being concerned about finding qualified staff in15052023.\14\ A skilled and trained workforce is vital for the stability1506and expansion of AI data centers--which rely on specialized data center1507technicians, network and electrical engineers, cybersecurity1508professionals, and project managers. CoreWeave supports efforts to1509develop a domestic workforce comprised of the skilled workers required1510to meet the growing AI demand and to accelerate AI innovation while1511creating the skilled good-paying jobs of the future.1512---------------------------------------------------------------------------1513    \14\ The White House, ``AI Talent Report,'' January 14, 20251514---------------------------------------------------------------------------1515Looking Forward: Ensuring Continued American Leadership in AI1516    The United States has built a remarkable lead in artificial1517intelligence through our unique combination of innovative and1518entrepreneurial private companies, world-class research institutions, a1519talented workforce, and a policy environment that fosters dynamic1520growth. This advantage is especially pronounced in AI infrastructure,1521where companies like CoreWeave have established global technological1522leadership in this critical layer of the AI stack.1523    However, the conditions that have enabled this position must be1524actively maintained while being flexible in order to adjust to new1525technological developments and political considerations. Countries1526around the world rightfully recognize the strategic importance of AI1527and are making coordinated efforts to build AI infrastructure. The1528decisions we make now will help determine whether America can maintain1529AI leadership in the years to come.1530    The current moment demands a thoughtful, transparent, and1531predictable approach that maintains our competitive edge, and seizes1532future opportunities while addressing legitimate concerns. As we1533consider policy options to address this dynamic sector, we should be1534attentive to how different approaches affect the entire AI ecosystem,1535from established players to new entrants, from model developers to1536infrastructure providers like CoreWeave.1537    In order to further America's lead in AI, we encourage the Federal1538Government to consider policies that:15391540   Foster a predictable investment environment through the1541        implementation of nationally consistent regulatory frameworks1542        for areas most critical to strengthen competitiveness and drive1543        innovation.15441545   Ensure that there are adequate, reliable supplies of power1546        at the lowest possible cost through policies which enhance the1547        ability to add generation and transmission to power next-1548        generation AI infrastructure. Careful reforms of existing1549        permitting and regulatory processes that enable affected1550        parties to participate in the process are needed for this to1551        occur.15521553   Maintain global competitiveness and strengthen U.S. industry1554        through strategically calibrated export policies that protect1555        national security while supporting the diffusion of the1556        American AI stack.15571558   Strengthen public-private partnerships, like the New Jersey1559        AI Hub, that accelerate innovation across research, industry,1560        and government.15611562    CoreWeave appreciates the Committee's leadership on these critical1563issues and we look forward to working with the Committee as it develops1564policies enabling the U.S. to maintain its leadership in the AI race.15651566    The Chairman. Thank you.1567    Mr. Smith.15681569 STATEMENT OF BRAD SMITH, VICE CHAIR AND PRESIDENT, MICROSOFT1570                          CORPORATION15711572    Mr. Smith. Chairman Cruz, Ranking Member Cantwell, members1573of the Committee, thank you for the opportunity to be here1574today. Let me just build on what my three colleagues have said1575and offer a few thoughts.1576    The first is I just wanted to refer to the chart on this1577easel that shows the AI tech stack that was also in my written1578testimony. It makes a simple but, I think, important point--we1579are all in this together.1580    If the United States is going to succeed in leading the1581world in AI it requires infrastructure. It requires success at1582the platform level. It requires people who create applications.1583    Interestingly, we at Microsoft get to work with all three1584of these leaders and their companies. Our success, each of our1585success, depends on each other's success, and what is true of1586the four of us is true when you look across the country and1587around the world at open source developers, people who are1588building power plants, electricians and pipe fitters who are1589going to work every single day.1590    So what do we need from the Congress and the country in1591order to succeed? I think it is three things. I described them1592in my written testimony.1593    First, as Chairman Cruz said, we need innovation.1594Innovation will go faster with, as Sam said, more1595infrastructure, faster permitting, more electricians.1596    We need more innovation fueled, as Ranking Member Cantwell1597said, by support from our universities and the Federal agencies1598that support basic research across the country, one of this1599country's crown jewels.1600    We also need, as Chairman Cruz said, faster adoption, what1601people refer to as AI diffusion--the ability to put AI to work1602across every part of the American economy to boost1603productivity, to boost economic growth, to enable people to1604innovate in their work, and the number-one ingredient for that1605history shows time and time again is skilling, investing in1606education.1607    And, finally, we need to export. If America is going to1608lead the world we need to connect with the world. We need to1609remember--I believe always--that as a country only 4 and a half1610percent of the world's people live in the United States of1611America.1612    Our global leadership relies on our ability to serve the1613world with the right approach to export controls and always,1614especially in technology, in our ability to sustain the trust1615of the rest of the world.1616    Ultimately, I think people who take the time--if they take1617the time to watch or read about this hearing may wonder what is1618this all about--what are we at this table trying to do? What do1619these two letters, AI, really mean to them?1620    Are we who are working in this industry trying to build1621machines that are better than people or are we trying to build1622machines that will help people become better?1623    Emphatically, it is and needs to be the latter. Are we1624trying to build machines that will outperform people in all the1625jobs that they do today or are we trying to build machines that1626will help people pursue better jobs and even more interesting1627careers in the future?1628    Indisputably, it needs to be the second, not the first, and1629I believe that is what we are and can do together.1630    As somebody who has now spent almost 32 years in this1631industry there are two things that always strike me. The first1632probably will not surprise you.1633    Never underestimate what technology can do, how quickly it1634can move, what it can accomplish. But the second is one that I1635think is too seldom discussed even though every day it stares1636us in the face.1637    Never estimate what people can do. Never underestimate1638human ambition. Never underestimate what a person can do if1639given a better technology tool and the ability to learn how to1640put it to use.1641    That is the story of this industry. It is the story of the1642country. It is, as you heard, the story of Sam Altman. Not1643everybody becomes a Sam Altman or a Satya Nadella or a Bill1644Gates but everybody deserves the opportunity to try.1645    Tonight across America, whether it is the attic of a house1646or the basement or just an everyday bedroom there are kids with1647computers, with phones, with access to the Internet and now the1648ability to put AI to work.1649    Let us invest in their education. Let us invest in the1650skills that the American public needs. Let us then invest in1651creating the future that the American public deserves.1652    Thank you.1653    [The prepared statement of Mr. Smith follows:]16541655      Prepared Statement of Brad Smith, Vice Chair and President,1656                         Microsoft Corporation1657    Chairman Cruz, Ranking Member Cantwell, and Members of the1658Committee,1659    Thank you for the opportunity to testify on the critical issue of1660artificial intelligence. I am Brad Smith, the Vice Chair and President1661of Microsoft Corporation.1662    AI has the potential to become the most useful tool for people ever1663invented. Like the general purpose technologies that preceded it, such1664as electricity, machine tools, and digital computing, AI will impact1665every part of our economy. It will shape not just how we work and live,1666but how we compete, prosper, and stay secure as a nation between now1667and the middle of this century.1668    The notice for this hearing aptly refers to an ``AI race.'' I would1669like to talk today about what is needed to win this race.1670    The AI race involves both technology and economics. It requires1671both innovation and diffusion. It is both a sprint and a marathon. The1672country can win a lap but lose the race if it fails to bring together1673all the ingredients needed for success.1674    It is a race that no company or country can win by itself.1675    To win the AI race, the United States will need to support the1676private sector at every layer of the AI tech stack. The nation will1677need to partner with American allies and friends around the world.1678    In my testimony today, I will focus on three strategic priorities1679where this Congress and the Federal government will make a difference.1680    First, the country must win the AI innovation race. This will1681require massive datacenters and AI infrastructure that need Federal1682support to expand and modernize the electrical grid on which they1683depend. The country must recruit and train skilled labor like1684electricians and pipefitters that are in short supply. We all must1685summon the best of our researchers at national labs and universities,1686supported by Federal basic research programs and partnerships that have1687become the envy of the world. We will need to continue to excel in1688moving innovative ideas from academic labs into companies and new1689products. And we will need to support AI developers with open and broad1690access to public data.1691    Second, the Nation must win the AI diffusion race. This will1692require that we promote broad AI adoption that will enable productivity1693growth across every sector of the economy. More than anything, this1694requires new initiatives to promote the AI skilling of the American1695workforce. This will involve basic AI fluency in our schools and new AI1696training programs in our community colleges. It will also include1697advanced AI education that will represent the next generation of1698computer science degrees, organizational skills that will be mastered1699in the country's business schools, and new courses in the nation's law1700schools. When combined, these will enable companies, non-profits, and1701government agencies alike to put AI to effective use. Governments at1702the federal, state, and local levels can then help accelerate this1703diffusion by adopting AI services to improve the effectiveness and1704efficiency of the services they provide to the public.1705    Third, the United States must export AI to American allies and1706friends. No company or country is so powerful that it can master the1707future of AI without friends. The United States and China are competing1708not only to innovate but to spread their respective technologies to1709other countries. This part of the race likely will be won by the1710fastest first mover. The United States needs a smart export control1711strategy that protects our national security while assuring other1712countries that they will have reliable and sustained access to critical1713American AI components and services. Perhaps as much as anything, this1714requires that we collectively sustain international trust in our1715products, our companies, and the country itself.1716                   AI as a General Purpose Technology1717    Economists sometimes put technologies into two categories, general1718purpose technologies and single-purpose tools. Most things in the world1719are single-purpose tools, like a smoke detector or a lawn mower. They1720do one thing very well. But over the course of history, certain so-1721called general purpose technologies impact and sometimes even redefine1722almost every sector of the economy. Electricity is the prototypical1723example, because when you think about it, electricity changed the way1724every economic sector works.1725    The key to mastering the future of AI starts in part by1726understanding the role technology has played in the past. The past1727three centuries have brought the world three industrial revolutions,1728each driven by these general purpose technologies. First, it was iron1729working in the United Kingdom, starting in the 1700s. And then it was1730electricity and machine tools in the 1800s, when the United States1731overtook the United Kingdom by putting these technologies to work more1732broadly than any other country. And then there was the third industrial1733revolution during the last 50 years, driven by computer chips and1734software.1735    Without question, being a global leader in advancing a general1736purpose technology gives a country a major edge. But one lesson of1737history is that the countries that benefit the most and advance the1738fastest are not necessarily the countries where the technology is1739invented. Rather, it's where the technology is diffused--or adopted--1740the most quickly and broadly. This is for good reason. If a technology1741improves productivity and changes every part of an economy, then the1742country that uses it the most broadly and quickly will benefit the1743most.1744    This both frames and defines the AI opportunity and challenge for1745the United States. As a nation, we need to focus both on advancing1746innovation and driving diffusion, both domestically and as a leading1747American export.1748                           The AI Tech Stack1749    The key to driving both innovation and diffusion is to recognize1750that AI, like all general purpose technologies, is built on what we in1751the industry call a tech stack--a stack of technologies that are used1752together. This is true for every great general purpose technology. You1753can see this, for example, if we go back in time and think about1754electricity. Thomas Edison first succeeded in 1878 in using electricity1755to light a lightbulb. But the illumination of lights across a city1756quickly required the construction of power plants, the fuel to run1757them, the creation of an electrical grid, the standardization of1758circuits, and a wide range of electrical appliances beyond the1759lightbulb itself. In short, a tech stack for electricity.1760    Artificial intelligence similarly is built on an AI tech stack.1761Fundamentally, it is divided into three layers, infrastructure, the1762platform layer, and applications. You can see this illustrated below.17631764[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]17651766    The infrastructure layer is massive. Microsoft is spending more1767than $80 billion this Fiscal Year on the capital investment needed for1768this layer, with more than half this amount being spent in the United1769States. This goes to buying land, investing in electricity and1770broadband connectivity, procuring chips like GPUs, and installing1771liquid cooling. These lead to the construction of datacenters--or often1772datacenter campuses with many buildings with potentially hundreds of1773thousands of computers. This infrastructure supports both the training1774of new AI models and their deployment, so they can be used for AI-based1775services around the world.1776    On top of this infrastructure, there is the platform layer. The1777heart of this layer consists of AI foundation models, including1778frontier models created by companies like OpenAI, as well as open1779source and other models from a wide variety of other firms--including1780Anthropic, Google, Mistral, DeepSeek, and Microsoft itself. The1781platform layer relies on data to train and ground models. And it1782includes a new generation of software-based AI platform services that1783are used to help build AI applications.1784    Ultimately, both the infrastructure and platform layers support the1785applications layer. These are devices and software applications that1786use AI to deliver better services to people. ChatGPT and Microsoft's1787Copilot are both examples of AI applications. One of the amazing things1788about the applications layer is it's not just companies--large or small1789or established or startup--that are creating AI applications. It's1790everybody. It's researchers using new AI-infused applications to change1791drug discovery. It's non-profits changing the way they deliver1792services. It's teachers using AI as a tool to improve the way they1793prepare material for a classroom. It's governments making everything1794from the filing of a tax return to the renewal of a driver's license1795easier and more efficient.1796    To build a new AI economy, it's critical to get all three of these1797layers working and to get a flywheel turning across the ecosystem. It's1798essential to build the infrastructure layer so people can develop and1799deploy the models at the platform layer. It's essential to use the AI1800models so that people will build the applications on top of them. And1801it's essential for customers to adopt the applications, so the market1802can grow, and drive increased investment to expand the infrastructure1803further. The process repeats itself. This is how a new economy is born.1804                  Success Requires an Entire Ecosystem1805    The flywheel effect makes clear that success requires not only1806national progress at one layer of the tech stack, but at every layer.1807That is what the private sector currently is pursuing in the United1808States better than in any other country. And it's what this Congress1809and the Executive Branch can help support with a strategy that promotes1810both AI innovation and diffusion up and down this stack.1811    National AI leadership requires not only success by a few1812companies, but by many. Today's panel, involving leading firms such as1813OpenAI, AMD, CoreWeave, and Microsoft, reflects important slices of the1814new AI economy. The AI economy requires a multifaceted and integrated1815ecosystem that includes ``Big Tech'' and ``Little Tech,'' startups and1816more established firms, open source and proprietary developers,1817suppliers and customers, firms that create data and firms that consume1818it, all working together. Governments as both regulators and leading AI1819adopters have critical roles to play.1820    Commentators sometimes focus on the tensions between different1821participants in this tech ecosystem. These deserve attention. What's1822often overlooked is that the different participants also depend on each1823other. And this means that the different contributors to the AI1824ecosystem all need to be healthy.1825    A large technology company like Microsoft has a unique1826opportunity--and responsibility--to partner with and support the1827participants at every level of the tech stack. We strive to advance not1828just innovation but an economic architecture, business models, and1829responsible practices that will help grow the AI market on a long-term1830basis. Not just for the United States, but the country's friends and1831allies.1832                      Winning the Innovation Race1833    Although the AI economy is being built mostly by the private1834sector, government policies and initiatives need to play a critical1835role. This starts with work needed to help fuel innovation. A few areas1836deserve particular attention in this hearing.1837Power the growth of datacenters1838    Just as you can't have reliable electricity in your home without a1839powerplant, you can't have AI without datacenters and AI1840infrastructure. And these datacenters require a vast supply chain to1841construct and large amounts of electricity to operate.1842    America's advanced economy relies on 50-year-old infrastructure1843that cannot meet the increasing electricity demands driven by AI,1844reshoring of manufacturing, and increased electrification. The United1845States will need to invest in more transmission and energy resources,1846onshore our supply chains, and modernize our electric grid to support1847forecasted increases in electrical loads. Microsoft is investing in1848these areas itself.1849    We urge the Federal government to streamline the Federal permitting1850process to accelerate growth in all these areas. The current Federal1851permitting processes often involve multiple agencies and complex,1852unpredictable, multi-year reviews. This hinders progress. The Federal1853government should take immediate steps to establish reliable,1854reasonable, and transparent timelines for permitting decisions. This1855can also be done by standardizing Federal permitting processes and1856designating a lead agency to shepherd the permits through the process.1857Further, the permitting agencies should utilize AI and digital tools to1858improve timelines and transparency for applicants and ensure the1859permitting agencies have quick access to information to assist them in1860their review and decision-making process.1861    We were pleased to see President Trump's recent Executive Order,1862``Updating Permitting Technology for the 21st Century,'' directing1863agencies to make maximum use of technology in the environmental review1864and permitting process. The Congress should also look to the Federal-1865State Modern Grid Deployment Initiative as a proven program that can be1866leveraged to deliver results.1867    This is just the start of what is needed to modernize and expand1868America's energy grid. We need to recognize that new investments in the1869grid are just as important today as they were a century ago, when the1870United States led the world in private and public sector support for1871electricity.1872Grow the AI Infrastructure workforce1873    Perhaps the single biggest challenge for data center expansion in1874the United States is a national shortage of people--including skilled1875electricians and pipefitters. Electricians, for example, are essential1876to datacenter construction, installing a complex system of electrical1877panels, transformers and backup power systems. We have hired thousands1878of electricians across the country, including in Arizona, Georgia,1879Virginia, Washington, and Wisconsin. But the United States doesn't have1880enough electricians to fill the growing demand. We estimate that over1881the next decade, the United States will need to recruit and train half1882a million new electricians to meet the country's growing electricity1883needs. We need a national strategy to ensure we meet this opportunity1884for American workers.1885    These are good jobs that will provide great long-term careers for1886people across the country. We recommend making existing Federal1887education and training funds, as well as tax incentives, available to1888scale up these opportunities. These could include targeting current1889Federal apprenticeship investments in regions that have identified1890major AI infrastructure initiatives and supporting existing training1891centers to quickly increase the number of registered apprenticeships1892focused on electricians.1893    We commend President Trump's recent Executive Order, ``Preparing1894Americans for High-Paying Skilled Trade Jobs of the Future,'' for1895highlighting the importance of skilled trades in the building of AI1896infrastructure and for paving the way to meet this moment. As Federal1897agencies work to implement the order, it will be critical that industry1898forecasters and union training centers work together to maximize1899impact.1900    Ultimately, we need new steps at every level of government and in1901communities across the country. For example, we need to do more as a1902nation to revitalize the industrial arts and shop classes in American1903high schools. This should be a priority for local school boards and1904state governments. Similarly, the nation's community colleges will need1905to do more to support a national initiative to help train a new1906generation of skilled labor, including electricians and pipefitters.1907Invest in AI research and development1908    To uphold America's position as a global scientific leader, it is1909imperative to enhance Federal investment in fundamental scientific1910research. The United States boasts a storied history of employing1911public-private partnerships. The decisions made decades ago to publicly1912fund research infrastructure and provide financial support to talented1913scientists and entrepreneurs paved a pathway to American technological1914leadership. Through federal, state and local government initiatives,1915investments were made in regional economies and programs, betting on1916the ingenuity of the American people. Notable incubators of the 20th1917century--such as Bell Labs and the network of Federal national1918laboratories--were the result of deliberate efforts to unite industry,1919government, and academia to propel scientific advancement. We must1920deploy a similar strategy today for AI and quantum technologies.1921Investments in these areas are critical to advancing the development of1922innovative technological solutions that address complex global1923challenges.1924    To outcompete nations like China, which have significantly boosted1925their research and development (R&D) investments, the United States1926must accelerate strategic investments in scientific research for future1927technologies. Experts predict China will continue to invest substantial1928resources in next-generation technologies such as AI, advanced1929manufacturing, clean energy, quantum computing, and semiconductors over1930the next decade.1931    Since the Second World War, America's technological innovation has1932been driven by R&D based on two critical ingredients that the rest of1933the world has both studied and envied. The first is sustained support1934for basic research. While a few tech companies invest substantial sums1935in basic research, as we do through Microsoft Research (MSR), most1936world-leading basic research is pursued by academics at American1937universities, often based on funding from the National Science1938Foundation and other Federal agencies. Driven by curiosity rather than1939a profit motive, this research often leads to unexpected but profound1940discoveries that are published publicly.1941    The second ingredient is a sustained commitment to investments in1942product development by companies of all sizes. The United States, more1943than any other country, has mastered the process of moving new ideas1944quickly from universities to the private sector. This success rests on1945healthy investments in both R and D, recognizing that basic research is1946often publicly funded and typically in universities, while product1947development is robustly and privately funded through companies. It's1948the combination of the two that makes American R&D so successful.1949    In 2019, President Trump approved an executive order designed to1950strengthen America's lead in artificial intelligence. It rightly1951focused on Federal investments in AI research and making Federal data1952and computing resources more accessible. Six years later, the President1953and Congress should expand on these efforts to support advancing1954America's AI leadership. More funding for basic research at the1955National Science Foundation and through our universities is one good1956place to start.1957Ensure public data is open and accessible1958    Data is the fuel that powers artificial intelligence. The quality,1959quantity, and accessibility of data directly determines the strength1960and sophistication of AI models. While the Internet has been a major1961source of training data, the Federal government remains one of the1962largest untapped sources of high-quality and high-volume data. Yet1963today, many of these datasets are either inaccessible or not usable for1964AI development.1965    By making government data readily available for AI training, the1966United States can significantly accelerate the advancement of AI1967capabilities, driving innovation and discovery. Opening access to these1968datasets would allow for the analysis of themes, patterns, and insights1969across broad datasets, propelling the country to the forefront of1970global AI development.1971    Importantly, accessible public data levels the playing field. It1972empowers not only large companies but startups, academic institutions,1973and nonprofits to train and refine AI models. This fosters a more1974competitive and inclusive AI ecosystem, where innovation is driven by1975ideas and ingenuity--not just proprietary data.1976    In comparison, countries like China and the United Kingdom are1977already investing heavily in their data resources, recognizing the1978economic and strategic value of national-scale data management. China's1979comprehensive system to manage datasets as a strategic resource and the1980UK's National Data Library underscore a growing global trend of1981treating data as a common good for economic competitiveness.1982                     Winning the AI Diffusion Race1983    History teaches us that the true impact of a general purpose1984technology is not measured solely by the caliber of its leading1985inventions, but by how quickly, widely, and effectively these are1986adopted across society. But the reality is that technology diffusion1987takes time, investment, partnerships, and sound public policy.1988    The history of electricity offers an important insight for AI. Once1989Thomas Edison proved in 1878 that electricity could power a lightbulb,1990why would anyone choose to sit at night in a room illuminated by a1991candle or kerosene? Yet tonight, almost 150 years later, more than 7001992million people on the planet still live without electricity in their1993homes. Diffusion requires not only great technology, but sound1994economics.1995    The economics of tech diffusion start with skilling. Countries need1996to invest in the skills needed to use new technology, both as1997individuals and across organizations. It is easy to underestimate both1998the role that skilling plays and the need for public policy to support1999it. But in each industrial revolution, the country that best harnessed2000the leading general purpose technology of its time was the Nation that2001skilled its population the most quickly and broadly.2002Skill the American workforce2003    In the new AI economy, Americans of all backgrounds will need2004critical AI skills to compete. To meet the totality of the skilling2005challenge, the country must pursue a new national goal to make AI2006skilling accessible and useful for every American. This will require a2007very broad range of partnerships and new policy ideas, spanning across2008geographic, organizational, economic, and political divides.2009    President Trump's recent executive orders focused on AI education2010and the workforce provide critical steps towards a national skilling2011strategy for AI. The ``Advancing Artificial Intelligence Education for2012American Youth'' EO establishes a clear policy to promote AI literacy2013by responsibly integrating AI into education for teachers and students.2014By fostering this early exposure, the Nation's youth will be better2015positioned for AI-enabled work. Congress can also consider leveraging2016existing Federal funding to the Nation's school districts to encourage2017AI learning and literacy in K-12 education.2018    Businesses and non-profits have important roles to play. At2019Microsoft, we are seeking to do our part to meet this skilling2020challenge. In 2025 alone, we are on a path to train 2.5 million2021Americans in basic AI skills. We're partnering with the National Future2022Farmers of America (FFA) to train educators in every state to integrate2023AI into the agricultural classroom through our Farm Beats for Students2024program. We are partnering with the American Federation of Teachers2025(AFT), the largest organization representing the Nation's educators in2026America, to deliver a co-developed training program to 10,000 AFT2027members. And we're partnering with the State of New Jersey, Princeton2028University, and CoreWeave on an AI Hub in New Jersey that will include2029support for AI education in local community colleges.2030    When it comes to AI skilling, the most important thing we need to2031do is recognize that this is a critical field that is ripe for2032attention, learning, partnership, and innovation. It will have a huge2033impact on broadening access to this technology across our economy and2034society. Generative AI is a new and young technology. So is our2035knowledge of the full extent of need in terms of AI skilling programs2036and support. This is a first-class priority that deserves as much2037attention and support as innovation in AI technology itself.2038Encourage AI adoption2039    The Federal Government also will play a critical role in AI2040diffusion by using AI itself. There are opportunities across the2041government to use AI to improve the quality and efficiency of public2042services for citizens.2043    It's encouraging to see the recent OMB publication of M-Memos2044focused on Federal government use and procurement of AI. Both memos2045emphasized the importance of removing barriers to innovation,2046maximizing the use of domestically developed AI products, and2047encouraging AI leaders within the Federal government to facilitate2048responsible AI adoption.2049    We're seeing activity in the states as well. We partnered with the2050Texas Department of Transportation to launch a six-week pilot program2051aimed at boosting productivity and improving decision-making across2052various departments. The program saw strong results with 97 percent of2053participants using the AI digital assistant during the pilot, 682054percent have integrated it into their daily workflow, and participants2055reporting saving an average of 12 hours a week on routine tasks.2056                         Exporting American AI2057    The ability to export our AI is essential to sustaining our global2058competitiveness and ensuring that our technological progress benefits2059not only our nation, but also our allies and partners around the world.2060Building on recent AI diplomacy efforts, the United States offers a2061compelling and trusted value proposition in the global technology2062landscape.2063    American tech companies, including Microsoft, are making2064unprecedented investments in AI infrastructure around the world.2065Microsoft alone is building AI infrastructure in more than forty2066countries, including regions where China has focused its investments.2067We urgently need a national policy that provides the right balance of2068export controls and trade support for these investments.2069    While the U.S. government rightly has focused on protecting2070sensitive AI components in secure datacenters through export controls,2071an even more important element of AI competition will involve a race2072between the United States and China to spread their respective2073technologies to other countries. Given the nature of technology markets2074and their potential network effects, this race between the United2075States and China for international influence likely will be won by the2076fastest first mover. The United States needs a smart international2077strategy to rapidly support American AI around the world.2078    This fundamental lesson emerges from the past twenty years of2079telecommunications equipment exports. Initially, American and European2080companies such as Lucent, Alcatel, Ericsson, and Nokia built innovative2081products that defined international standards. But as Huawei invested2082in innovation and China's government subsidized sales of its products,2083especially across the developing world, adoption of these Chinese2084products outpaced the competition and became the backbone of numerous2085countries' telecommunications networks. This created the technology2086foundation for what later became an important issue for the Trump2087Administration in 2020, as it grappled with the presence of Huawei's 5G2088products and their implications for national and cybersecurity.2089    Early signs suggest the Government of China is interested in2090replicating its successful telecommunications strategy. China is2091starting to offer developing countries subsidized access to scarce2092chips, and it's promising to build local AI datacenters. The Chinese2093wisely recognize that if a country standardizes on China's AI platform,2094it likely will continue to rely on that platform in the future.2095    International partnerships will be critical. This is why Microsoft2096has partnered with entities like the UAE's G42 and investment funds2097like Blackrock and MGX, aiming to raise up to $100 billion for AI2098infrastructure and supply chains. American tech companies and private2099capital markets are forging stronger ties with key nations and2100sovereign investors in the Middle East, surpassing previous efforts to2101counter Chinese subsidies in telecommunications and reflecting our2102commitment to innovation and cooperation. While China's government may2103subsidize its technology adoption in developing regions, it will2104struggle to match the scale and impact of America's private sector2105investments.2106    Pragmatic American export control policies are essential, balancing2107security protections with the ability to expand rapidly. Protecting2108national security by preventing adversaries from acquiring advanced AI2109technology is crucial. Rules should include qualitative standards for2110secure datacenter deployments to prevent chip diversion to China and2111ensure advanced AI services are safeguarded. We support this type of2112approach.2113    However, we have expressed our concerns about the quantitative caps2114imposed on GPU shipments by the interim final AI Diffusion Rule issued2115in January. These place key American allies and partners in a Tier Two2116category, imposing limits on AI datacenter expansion. This includes2117countries like Switzerland, Poland, Greece, Singapore, India,2118Indonesia, Israel, the UAE, and Saudi Arabia. Customers in these2119countries now fear restricted access to American AI technology--2120potentially benefitting China's AI sector by turning to alternatives.2121    The Trump administration has an opportunity to revise the rule,2122eliminating quantitative caps and retaining qualitative standards. This2123approach ensures American allies and partners remain confident in2124accessing American AI products.2125    Ultimately, we need to recognize that countries around the world2126will use American AI only if they can trust it. This creates2127responsibilities for American companies to develop and deploy AI2128infrastructure and products in a responsible manner that meets local2129needs. And it requires that countries have confidence in sustained and2130uninterrupted access to critical AI components and services. The United2131States has long built a reputation for trustworthy technology that2132China has been unable to match. But this reputation, like everything2133that truly matters, requires constant attention and care.21342135                 STATEMENT OF HON. TIM SHEEHY,2136                   U.S. SENATOR FROM MONTANA21372138    Senator Sheehy [presiding]. Thank you, witnesses, for your2139testimony and I will start off with the first round of2140questions and move down dais to our Ranking Member here.2141    Thank you for your testimony. It certainly makes me sleep2142better at night worried about Terminator and Skynet coming2143after us, knowing that you guys are behind the wheel.2144    But in five words or less, starting with you, Mr. Smith,2145what are the first--what are the five words you need to see2146from our government to make sure we win this AI race?2147    Mr. Smith. More electricians. That is two words.2148    [Laughter.]2149    Mr. Smith. Broader AI education.2150    Senator Sheehy. And no using ChatGPT as a phone a friend.2151    Mr. Intrator. Thank you. I would say that we need to focus2152on streamlining the ability to build large things.2153    Ms. Su. Policies to help us run faster in the innovation2154race.2155    Mr. Altman. Allow supply chain. Sensible policy.2156    Senator Sheehy. That was good. So what I hear there is2157something pretty similar to the races we have won before--2158nuclear energy, for example.2159    You know, the Germans and Austrians really led the2160innovation around that but we won the race because we put a2161massive government effort, collaborating with our universities2162and others to win that race.2163    Space--you know, the Soviets put the first satellite up,2164put the first man in space, but we won the space race because2165we adopted a framework to ensure that we won. Aviation,2166automobiles, et cetera.2167    So what I hear from you is you do need support from our2168government but you also need the government to stay out of your2169way so you can innovate and win this race.2170    How do we incentivize companies to do business here in2171America to make sure we win this race in America and America2172leads not just China but other nonstate actors, too?2173    I mean, I think that the scariest thing about AI from a2174capability standpoint is it does not have to be a state actor2175to win this race. It is not like nuclear energy. It is not like2176space technology. A nonstate actor could just as easily win2177this race and wield more power than anyone else.2178    So how do we encourage innovators' investment to happen2179here in America to ensure we win this race?2180    Mr. Altman, do you want to start?2181    Mr. Altman. We were honored to announce back in January,2182Project Stargate, a $500 billion investment in United States2183infrastructure. That is now well underway, as I mentioned,2184getting to see it yesterday in Abilene. The first site was2185incredible.2186    We need a lot more of that. We need certainty on the2187ability to build out this entire supply chain, build the data2188centers, permit the electricity. We would love to bring chip2189production here, network production here, server rack2190production here.2191    And I think the world does want to invest. We have a lot of2192global investment flowing into the U.S. to do this. We also2193want to make sure that other countries are able to build with2194our technology, use our models, and sort of, like, be in our2195orbit and, you know, use U.S. diffusion of technology here.2196    So that is really important. We need to make sure that the2197highest skilled researchers that want to come work at U.S.2198companies can come here and do that. We need to make sure that2199companies like OpenAI and others have legal clarity on how we2200are going to operate.2201    Of course, there will be rules. Of course, there need to be2202some guardrails. This is a very impactful technology. But we2203need to be able to be competitive globally. We need to be able2204to train.2205    We need to be able to understand how we are going to offer2206services and sort of where the rules of the road are going to2207be, so clarity there. And I think an approach like the2208internet, which did lead to flourishing of this country in a2209very big way, we need that again.2210    Senator Sheehy. Dr. Su.2211    Ms. Su. I would add I think computing is a foundation to2212all of this. We want to have more compute built in the U.S. by2213U.S. companies and ensure that we have a great environment for2214that. We want to ensure that our technology around the world is2215also used broadly and in the right ways.2216    So I think the conversation about export controls and rules2217should just be simple, easy to follow, easy to enforce, and2218enable U.S. AI platforms to be the foundation.2219    And then, certainly, the comments around bringing2220manufacturing back home and ensuring that we have the right2221talent base are all extremely important elements of that.2222    Senator Sheehy. Are companies weighing doing business in AI2223in America versus China? Are the companies making that side by2224side comparison?2225    Ms. Su. I think if you look across the world there are2226countries and companies that will ask those questions. You2227know, if it is hard to obtain U.S. technology--although U.S.2228technology is the best, if it is hard to obtain then there is a2229hunger for AI and they will choose what is available and if2230China is available that will certainly be a outcome that we2231would not like to see.2232    Senator Sheehy. Well, I think I hear the words2233infrastructure, electricians, universities, regulatory2234framework, and I think those are things we can help with.2235    I hear words like innovation and talent and I say--I hear2236Dr. Su, run faster. Those are not things--the government cannot2237manufacture talent. We cannot make you run faster.2238    But we can give you the tools to do that and I think it is2239time that we create a framework so that you have the tools you2240need to win this race because you are going to be the ones that2241win it, not us.2242    Thank you for your testimony.2243    Ranking Member Cantwell.2244    Senator Cantwell. Thank you, Mr. Chairman. I would like to2245continue that same theme generally about competitiveness. Do we2246need NIST to set standards?2247    If you could just yes or no and just go down the line.2248    Mr. Altman. I do not think we need it. It can be helpful.2249    Ms. Su. Yes.2250    Mr. Intrator. Yes.2251    Mr. Smith. Yes.2252    Senator Cantwell. OK. So in the context of what we are2253talking about here, we are really just talking--I do not know,2254Mr. Smith or Mr. Intrator or Dr. Su, any.2255    The issue here is if we want to move fast we want to create2256just like with electricity the standards by which we want to2257move fast.2258    Here, I would just call it code for code is what we want,2259right? We want NIST to do something in the standard setting2260that will allow us to move much faster. Is that right?2261    Either Mr. Smith or Mr. Intrator.2262    Mr. Smith. What I would say is this.2263    First of all, NIST is where standards go to be adopted but2264it is not necessarily where they first go to be created. So we2265have got what----2266    Senator Cantwell. Right. Thank you for that clarity. We are2267talking about a industry IEEE, you know, lots of different2268organizations, industry input, and then they are adopted.2269    So yes, let us clarify that. Let us clarify that.2270    Mr. Smith. Yes. I think that is the way it works.2271    Senator Cantwell. Yes. But you think we need to do that,2272particularly if the United States wants to lead?2273    Mr. Smith. We will need industry standards. We will need2274American adoption of standards. And you are right, we will need2275U.S. efforts to really ensure that the world buys into these2276standards.2277    Senator Cantwell. OK. Mr. Intrator.2278    Mr. Intrator. I think it is important that when you are2279working with standards what that allows for is a common2280vocabulary which allows for acceleration.2281    And so to the extent that we can step into that role and2282establish touch points where everyone can agree on specific2283things that will lead to an acceleration both domestically and2284abroad.2285    Senator Cantwell. And I do not know if, you know, drilling2286down more on what you think those are but in general, you know,2287when I think about the Internet and HTTP or HTML or any of2288the--TCPIP--we are talking about things that allowed us to move2289faster and getting those standards established helped us do2290that.2291    On the export issue, Mr. Intrator, the issue of cloud2292sources should not be left out. If we say let us go with2293Malaysia, Malaysia is going to tell us that they can certify2294that there is no, you know, diversion of these ships to--you2295know, to China and we basically have a way that we can make2296sure that this is understood and monitored then we also want2297access, right? We want access by U.S. companies.2298    Mr. Intrator. Yes. I think Lisa's point was excellent,2299right? At the end of the day, the world wants to be able to2300build and deploy artificial intelligence in a very broad way2301and if we--you know, nature abhors a vacuum.2302    If we do not step into that role other technology will step2303in that role. If it is suboptimal so be it. It is better to2304have something that is suboptimal than have nothing and so that2305is what----2306    Senator Cantwell. Well, we do not want a reoccurrence of a2307Huawei that develops faster and then has a government back door2308and then we all have to raise opposition. I am for a tech NATO.2309    I am for the five most sophisticated democracies and tech2310nations setting the rules of the road and saying, this is who2311you should buy from. Do not buy from anybody else who has a2312government back door. Not a good idea.2313    So that is how we get leverage. You know, I am not so hot2314on the President's tariff agenda for this very reason because2315we are not building the alliances.2316    We are creating the enemies, and what I want to do is get2317the supply chain here, get the semiconductor flow here, lower2318the cost, and go as fast as we can.2319    Mr. Intrator. Yes, I agree with that. I do not think that2320that is--I do not think anybody is not going to agree with2321that, right? I think that is an excellent objective.2322    I just think that what will happen beyond the five NATO2323companies is that there will be a demand for artificial2324intelligence and they will proceed with what they can proceed2325with.2326    Senator Cantwell. Dr. Su, what is your view of this about2327how we win, how we protect our objectives, but we are more2328aggressive on the export strategy?2329    Ms. Su. Well, I think there is a clear recognition that we2330need an export strategy and so having--you know, having this2331conversation is very important, and from our perspective the2332idea is to ensure that our allies--and, frankly, I use allies2333in the very broadest sense--get access to the great American2334technology that we have with the appropriate controls in place,2335and I think you can do both.2336    To your earlier comment, Ranking Member Cantwell, about the2337need to have U.S. technologies in those countries I think those2338countries are actually very interested in doing that because we2339do have the best technology today, and using that to really2340build this broad AI ecosystem is really our opportunity.2341    Senator Cantwell. I agree. Thank you so much.2342    Senator Sheehy. The senior senator from Ohio.23432344               STATEMENT OF HON. BERNIE MORENO,2345                     U.S. SENATOR FROM OHIO23462347    Senator Moreno. Thank you, Chairman Sheehy. Make sure2348Senator Cruz heard that one.2349    So, first of all, thank you for being here and taking the2350time. If I could just real quickly just confirm that I have2351heard what you said pretty unanimously, which is we need2352dramatically more power generation in this country. Is that2353correct?2354    All right. So, Dr. Su, you just recently did a partnership2355with TSMC to manufacture your chips here in America. Thank you.2356I think it is a little bit long overdue. I wish you had--we had2357done more of that earlier.2358    Are those semiconductor fabs high energy users?2359    Ms. Su. Thank you, Senator.2360    We are very pleased with our efforts together with the2361government on bringing more manufacturing back to the United2362States.2363    To your question, certainly, semiconductor manufacturing2364plants are high energy users and we do need more power for both2365manufacturing as well as for data centers, as you mentioned.2366    Senator Moreno. And without chips this just does not work.2367Like, if we do not have the highest performance chips made here2368in the United States this is not going to happen here, correct?2369    Ms. Su. We absolutely need the highest performing chips and2370we also need the entire ecosystem for chip manufacturing. So2371wafers are one piece but there are many other pieces as well.2372    Senator Moreno. And are those chips powered by solar power2373and windmills?2374    Ms. Su. Today they are not but I think there are2375opportunities to certainly do that.2376    Senator Moreno. So do you think it is outrageous that last2377year because of the policies of the Biden administration that237890 percent of new power generation in this country was2379windmills and solar panels and we absolutely kneecapped2380American energy?2381    We have a thousand years of natural gas sitting in2382Pennsylvania, Ohio, and West Virginia, and yet 90 percent of2383power generation in this country last year was solar panels and2384windmills. Does that make this country more competitive or less2385competitive?2386    Anybody can jump into that one that wants to answer that.2387    Mr. Smith. Let me say two things. One, you are right, we2388need more electricity. I think our industry, it is worth2389remembering, is only going to account for 15 percent of the2390total additional electricity the country is going to need.2391    We are going to need electricity from a variety of sources.2392Today in the United States 56 percent of our electricity comes2393from carbon. Forty-four percent comes from carbon-free energy,2394meaning nuclear wind, or solar. We need a broad based approach2395and we need a diversity of sources.2396    Senator Moreno. And, again, 90 percent was energy that is2397not affordable, it is not abundant, and it is not reliable.2398    Let me just shift gears. Mr. Altman, thank you for, first2399of all, creating your platform in an open basis and agreeing to2400stick to the principles of nonprofit status. I think that is2401very important.2402    Do you think that the Internet age did a good job between2403the beginning of the 1990s through the 2000s of protecting2404children?2405    Mr. Altman. I would say not particularly.2406    Senator Moreno. Yes. And you are a new father, correct?2407    Mr. Altman. Yes.2408    Senator Moreno. Congratulations.2409    Mr. Altman. Thank you very much.2410    Senator Moreno. He is doing well?2411    Mr. Altman. He is. It is the most amazing thing ever.2412    Senator Moreno. Yes. I do not think you want your best--2413your child's best friend to be an AI bot.2414    Mr. Altman. I do not.2415    Senator Moreno. So what can we do? How can we work together2416to protect children?2417    Mr. Altman. We have talked a lot about some of the things2418we are doing here. We are trying to learn the lessons of the2419previous generation and, you know, that is kind of the way it2420goes. People make mistakes and you do it better next time.2421    One thing we say a lot internally is we want to treat our2422adult users like adults. We want to give them a lot of2423flexibility.2424    We want to let them use the service with a lot of freedom,2425and for children there needs to be a much higher level of2426protection, which means the service will not do things that2427they might want.2428    Now, we are still early so sometimes people say, oh, you2429are being too strict on the rules and it is just we cannot2430perfectly, like, tell this.2431    But if we could draw a line and if we knew for sure when a2432user was a child or an adult we would allow adults to be much2433more permissive and we would have tighter rules for children.2434    Senator Moreno. So I think what I would ask is if you could2435have your team commit to having your teams work with our teams2436to make certain that we put together the right framework early2437on I think is the best way we can move forward, because we do2438not want to overregulate but we cannot repeat the mistakes of2439the Internet and social media era where children got harmed.2440    Mr. Altman. We would be delighted to work with you all. I2441think it is super important.2442    Senator Moreno. Thank you.2443    Mr. Altman. Can I say one more thing about what you said?2444    Senator Moreno. Of course.2445    Mr. Altman. This idea of AI and social relationships I2446think this is a new thing that we need to pay a lot of2447attention to.2448    People are relying on AI more and more for life advice,2449sort of, emotional support, that kind of thing. It is a newer2450thing in recent months, but I--and I do not think it is all2451bad, but I think we have to, like, understand it and watch it2452very carefully.2453    Senator Moreno. Thank you, and thank you for that2454commitment. It is very appreciated. I have talked to your team2455already. Good people.2456    Mr. Altman. Great.2457    Senator Moreno. Mr. Intrator, real quickly, can you talk2458about the intersection between the importance of a robust2459stablecoin ecosystem here in America and how that has a future2460with payments and how AI will factor into that? Because I do2461not think people see how this fits into the broader puzzle.2462    Mr. Intrator. So thank you for the question.2463    And we did start out as a crypto-based company hobby that2464kind of got away from us a little bit.2465    Look, I think that stablecoins, crypto, AI, they share2466certain DNA in common which is that they are attempts to build2467into a future where new technology will make things better for2468society and there is a huge potential for us to use2469stablecoins, crypto, and AI in a combination for better2470outcomes.2471    Senator Moreno. All right. Thank you.2472    And that was the quickest coup since 1959.2473    [Laughter.]2474    The Chairman [presiding]. Senator Klobuchar.24752476               STATEMENT OF HON. AMY KLOBUCHAR,2477                  U.S. SENATOR FROM MINNESOTA24782479    Senator Klobuchar. Thank you very much, Senator Cruz.2480    A lot of exciting things with AI, especially from a state2481like mine that is home to the Mayo Clinic with the potential to2482unleash scientific research while we have mapped the human2483genome and we have rare diseases that can be solved.2484    So there is a lot of positive but we all know, as you have2485all expressed, there are challenges that we need to get at with2486permitting reform. I am a big believer in that. energy2487development.2488    Thank you, Mr. Smith, for mentioning this with wind and2489solar and the potential for more fusion and nuclear, but wind2490and solar the price going down dramatically in the last few2491years and to get there we are going to have to do a lot better.2492    I think David Brooks put it the best when he said, ``I2493found it incredibly hard to write about AI because it is2494literally unknowable whether this technology is leading us to2495heaven or hell''.2496    We want it to lead us to heaven and I think we do that by2497making sure we have some rules of the road in place so it does2498not get stymied or set backward because of scams or because of2499use by people who want to do us harm.2500    As mentioned by Senator Cantwell, Senator Thune and I have2501teamed up on legislation to set up basic guardrails for the2502riskiest nondefense applications of AI.2503    Mr. Altman, do you agree that a risk-based approach to2504regulation is the best way to place necessary guardrails for AI2505without stifling innovation?2506    Mr. Altman. I do. That makes a lot of sense to me.2507    Senator Klobuchar. OK, thanks. And did you figure that out2508in your attic?2509    Mr. Altman. No, that was a more recent discovery.2510    Senator Klobuchar. Thank you. Very good. Just want to make2511sure.2512    Our bill directs, Mr. Smith, the Commerce Department to2513develop ways of educating consumers on how to safely use AI2514systems. Do you agree that consumers need to be more educated?2515This was one of your answers to your five words so I assume you2516do.2517    Mr. Smith. Yes, and I think it is incumbent upon us as2518companies and across the business community to contribute to2519that education as well.2520    Senator Klobuchar. OK, very good.2521    Back to Mr. Altman. Americans rely on AI, as we know,2522increasingly on some high impact problems. To make them be able2523to trust that we need to make sure that we can trust the model2524output.2525    The New York Times recently reported earlier this week that2526AI hallucinations--a new word to me--where models generate2527incorrect or misleading results are getting worse. That is2528their words.2529    What standards or metrics does OpenAI use to evaluate the2530quality of its training data and model outputs for correctness?2531    Mr. Altman. On the whole, AI hallucinations are getting2532much better. We have not solved the problem entirely yet, but2533we have made pretty remarkable progress over the last few2534years.2535    When we first launched ChatGPT it would hallucinate things2536all the time. This idea of robustness, being sure you can trust2537the information, we have made huge progress there. We cite2538sources.2539    The models have gotten much smarter. A lot of people use2540these systems all the time and we were worried that if it was2541not 100.0 percent accurate, which is still a challenge with2542these systems, it would cause a bunch of problems.2543    But users are smart. People understand, you know, what2544these systems are good at, when to use them, when not, and as2545that robustness increases, which it which it will continue to2546do, people will use it for more and more things.2547    But we have made--as an industry we have made pretty2548remarkable progress in that direction over the last couple of2549years.2550    Senator Klobuchar. I know we will be watching that. Another2551challenge that has been--we have seen, and Senator Cruz worked2552and I worked on a bill together for quite a while and that is2553the Take It Down Act, and that is that we are increasingly2554seeing Internet activity where kids looking for a boyfriend or2555a girlfriend, maybe they put out a real picture of themselves.2556    It ends up being distributed at their school or they2557somehow they--someone tries to scam them for financial gain, or2558it is AI, as we have increasingly seen where it is not even2559someone's photos but someone puts a fake body on there and we2560have had about over 20 suicides in one year of young people2561because they felt like their life was ruined because this was--2562they were going to be exposed in this way.2563    So this bill we passed and through the Senate and the2564House. The First Lady supported it and it is headed to the2565President's desk. Could you talk about how we can build models2566that can better detect harmful deep fakes, Mr. Smith?2567    Mr. Smith. Yes. I mean, we are doing that. OpenAI is doing2568that. A number of us are and I think the goal is to first2569identify content that is generated by AI and then often it is2570to identify what kind of content is harmful, and I think we2571have made a lot of strides in our ability to do both of those2572things.2573    There is a lot of work that is going on across the private2574sector and in partnership with groups like NCMEC to then2575collaboratively identify that kind of content so it can be2576taken down.2577    We have been doing this in some ways for 25 years since the2578Internet and we are going to need to do more of it.2579    Senator Klobuchar. And on the issue--last question, Mr.2580Chair. Since the last one was about your bill I figured it is2581OK. The newspapers, and you testified before the Senate2582Judiciary Committee, Mr. Smith, about the bill.2583    Senator Kennedy and I still think that there is an issue2584here about negotiating content rates. We have seen some action2585recently in Canada and other places.2586    Can you talk about those evolving--the dynamics with AI2587developers and what is happening here to make sure that content2588providers and journalists get paid for their work?2589    Mr. Smith. Yes, it is a complicated topic but I will just2590say a couple of things.2591    First, I think we should all want to see newspapers in some2592form flourish across the country including, say, rural counties2593that increasingly have become news deserts. Newspapers have2594disappeared.2595    Second, and it has been the issue that we discussed in the2596Judiciary Committee, there should be an opportunity for2597newspapers to get together and negotiate collectively. We have2598supported that. That will enable them to basically do better.2599    Third, every time there is new technology there is a new2600generation of a copyright debate. That is taking place now.2601Some of it will probably be decided by Congress, some by the2602courts.2603    A lot of it is also being addressed through collaborative2604action, and we should hope for all of these things to, I will2605just say, strike a balance. We want people to make a living2606creating content and we want AI to advance by having access to2607data.2608    Senator Klobuchar. OK, thanks. I will ask other questions2609on the record. Thank you, Mr. Chair.2610    The Chairman. Thank you.2611    You know, Senator Klobuchar asked whether AI will lead us2612to heaven or hell. It reminded me of a famous observation by2613Yale Law Professor Grant Gilmore that in heaven there is no law2614and the lion will lie down with the lamb. In Hell there is2615nothing but law and due process is meticulously observed.2616    Let me ask you this, and this is to each of the four2617witnesses. In the race for AI who is winning, America or China?2618If the answer is America how close is China to us, and what do2619we do to make sure the answer remains America will win?2620    Mr. Altman, we will start with you.2621    Mr. Altman. It is our belief that the American models,2622including some models from our company OpenAI and Google and2623others are the best models in the world.2624    It is very hard to say how far ahead we are but I would say2625not a huge amount of time, and I think to continue that2626leadership position and the influence that comes with that and2627all of the incredible benefits of the world using American2628technology products and services, the things that my colleagues2629have spoken about here, the need to win in infrastructure,2630sensible regulation that does not slow us down, the sort of2631spirit of innovation and entrepreneurship that I think is a2632uniquely American thing in the world, none of this is rocket2633science.2634    We just need to keep doing the things that have worked for2635so long and not make a silly mistake.2636    The Chairman. Dr. Su.2637    Ms. Su. I will answer in the realm of chips. I would say2638America is ahead in chips today. We have the best AI2639accelerators in the world.2640    I think China, although they have restrictions, given their2641ability to use advanced technologies, the one thing that is2642very important for us all to remember is there are multiple2643ways to do things. You know, having the best chips is great,2644but even if you do not have the best chips you can get a lot2645done.2646    So I think this conversation about how far behind China is2647they are certainly catching up because there are many ways to2648do things.2649    I think relative to what we can do I will continue to say2650really ensure that our spirit of innovation is allowed to work2651and that is having very supportive government policies to do2652that, having very consistent policies and allowing us to do2653what we do best, which is innovate at every layer of the stack.2654    The Chairman. Mr. Intrator.2655    Mr. Intrator. So I will speak to it from the physical2656infrastructure and software stack to deliver that.2657    America is ahead, but it is the Achilles heel from the2658perspective of the ability, as I started to--better? Sorry2659about that.2660    So the ability to build very large solutions to the2661computing infrastructure component of this is an area that we2662are going to struggle with from a permitting and building large2663projects to be able to deliver the power to allow those2664building artificial intelligence to continue to move as fast as2665they can in the race that we are in.2666    The Chairman. Mr. Smith.2667    Mr. Smith. I think the United States has a lead today in2668what is a close race and a race that will likely remain close.2669    The number-one factor that will define whether the United2670States or China wins this race is whose technology is most2671broadly adopted in the rest of the world.2672    This is a global market and it will be defined as2673technology markets typically are by network effects. Eighteen2674percent of the people of the world live in China. Four percent2675live in the United States. Seventy-eight percent live somewhere2676else.2677    The lesson from Huawei and 5G is whoever gets there first2678will be difficult to supplant. We need to export with the right2679kinds of controls. We need to win the trust of the rest of the2680world.2681    We need to have the financial architecture that gets not2682only to the countries that are industrialized but the nations,2683say, across Africa where typically China and Huawei have done2684so well.2685    The Chairman. So some of my colleagues have made reference2686to standards as something that is desirable, and I will say2687standards is often code word for regulations and, indeed, the2688EU stifling standards concerning the Internet is what killed2689tech in Europe.2690    We are seeing now state legislatures mimicking the EU such2691as California's S.B. 1047 which, thankfully, was overwhelmingly2692defeated but would have created essentially a California DMV2693for AI model registration.2694    How harmful would it be to winning the race for AI if2695America goes down the road of the EU and creates a heavy handed2696prior approval government regulatory process for AI?2697    Mr. Altman. I think that would be disastrous. To give a2698more specific answer to your previous question, which I think2699touches on why it would be so bad, there are three key inputs2700to these AI systems.2701    There is compute, all the infrastructure we are talking2702about, there is algorithms that we all do research on, and2703there is data.2704    If you do not have any one of those you cannot succeed in2705making the best models and, as Brad said, the way for America2706to influence the world here is to have the technology that2707people most want to use and most adopt.2708    The world uses iPhones and Google and Microsoft products,2709and that is wonderful. Like, that is how we have our influence.2710We do not want that to stop happening. So systems that stop us2711on any of these areas, you know, if we have rules about what2712data we can train on that are not competitive with the rest of2713the world then things can fall apart.2714    If we are not able to build the infrastructure and2715particularly if we are not able to manufacture the chips in2716this country the rules can fall apart. If we cannot build the2717products that people want that naturally win in the market--and2718I think people do want to use American products.2719    We can make them the best. But if we are prevented from2720doing that people will use a better product made from somebody2721else that does not have the sort of--you know, that is not2722stymied in the same way.2723    So it is--I am nervous about standards being set too early.2724I am totally fine, you know, at the position that some of my2725colleagues took that standards, once the industry figures out2726what they should be, it is fine for them to be adopted by a2727government body and sort of made more official.2728    But I believe the industry is moving quickly toward2729figuring out the right protocols and standards here and we need2730the space to innovate and to move quickly.2731    The Chairman. So if each of you could briefly answer that2732question because my time has expired. So I want to be2733respectful of that.2734    Ms. Su. I agree with the comments that Sam put up.2735    Mr. Smith. I agree, and I would just say and I think the2736point you are making is we have to be very careful not to have2737these preapproval requirements including at state levels2738because that would really slow innovation in the country.2739    Mr. Intrator. I think that a patchwork of regulatory2740overlays will cause friction in the ability to build and extend2741what we are doing.2742    The Chairman. Thank you.2743    Senator Curtis? Schatz?2744    Senator Schatz? Apologies.27452746                STATEMENT OF HON. BRIAN SCHATZ,2747                    U.S. SENATOR FROM HAWAII27482749    Senator Schatz. No problem, Chairman.2750    Thank you for being here. I just want to follow up on the2751Chairman's question and a sort of--maybe an emerging consensus2752on the Committee.2753    OK. I do not think there is anybody even on this side of2754the dais that is proposing a sort of European style2755preapproval.2756    I think there are some people who would like to do nothing2757at all in the regulatory space but I think most people2758understand that some guardrails--those are the words that you2759use, Mr. Altman--rules and guardrails are necessary.2760    Are you saying that self-regulation is sufficient at the2761current moment?2762    Mr. Altman. No. I think some policy is good. I think it is2763easy for it to go too far, and as I have learned more about how2764the world works I am more afraid that it could go too far and2765have really bad consequences.2766    But people want to use products that are generally safe.2767You know, when you get on an airplane you kind of do not think2768about doing the safety testing yourself.2769    You are, like, this is--well, maybe this is a bad time to2770use the airplane example but you kind of like want to just2771trust that you can get on an----2772    Senator Schatz. It is an excellent time to use the airplane2773example. But I think your point is exactly right is that, look,2774there is a race but we need to understand what we are racing2775for, right?2776    And it also has to do with American values. It is not just2777a sort of commercial race so we can edge out our near peer2778competitor both in the public sector and the private sector. We2779are trying to win a race so that American values prevail2780internationally.2781    Mr. Smith, I want to move on to another topic. It seems to2782me that on the consumer side that one of the most basic rights2783of a user on the Internet is to understand what they are2784looking at or listening to and whether or not it was created2785solely by a person, a person using an AI, or automatically2786generated using AI.2787    Do you think a labeling regime--not a prohibition on the2788use of AI but just the disclosure, especially as it relates to2789images, music, creativity--do you think a label would be2790helpful for consumers?2791    Mr. Smith. Generally, yes, and I think that is what we in2792the industry have been working to create. I think you are right2793to make the distinction and focus especially on, say, images,2794video, audio files.2795    There is a standard called C2PA that we and a number of2796companies now have been advancing. It has content credentials.2797It enables people to know where something was created, who2798created it, and I think--you are right--to know whether it was2799created by a person, by AI, or a person with the help of, say,2800AI.2801    Senator Schatz. I just want to use sort of common language,2802not the language that all of you use or that we have all2803learned to use.2804    When you talk about the data as one of the three elements2805that makes a model work, data really is intellectual property.2806It is human innovation, human creativity, and I do think we may2807have a disagreement--and I agree with Senator Klobuchar about2808the need to understand that these models have been trained on2809data but what we are really talking about is human achievement2810all the way up to now.2811    And I have a deep worry--look, I am actually an optimist in2812the energy space and the public service space, certainly in2813health innovation. There are a lot of really exciting2814opportunities here.2815    But we got to pay people for their knowledge and I am2816concerned that these models are going to be so successful in2817spitting out what appears to be knowledge that we are going to,2818on the back end, not pay people for all of the inputs and we2819will have a sort of stalling out of these models.2820    And you talked about a tension but I am trying to figure2821out what the tension really is other than you would like to pay2822as little as possible for these inputs.2823    Go ahead, Mr. Smith.2824    Mr. Smith. Well, you had me until the last sentence.2825    Senator Schatz. I know.2826    Mr. Smith. Look, we create intellectual property. We2827respect intellectual property. So we are emphatically of the2828view that intellectual property and the creation of it should2829be rewarded.2830    Ultimately, intellectual property laws are always about2831drawing the line. It is really the line that you refer to. In2832copyright, there is expression that is protected. If you write2833a book and somebody copies it then you are entitled to be paid.2834    But there are ideas. If someone reads your book, if someone2835remembers that Shakespeare wrote a story about two teenagers2836who fell in love----2837    Senator Schatz. Sure. Then that is fair use.2838    Mr. Smith [continuing]. Then that is fair use. That is why2839this country and Congress created it.2840    Senator Schatz. OK. That is where the tension is.2841    Mr. Smith. That is what we need to focus on.2842    Senator Schatz. With your permission, Chairman, I want to2843ask one final question.2844    The Chairman. Proceed.2845    Senator Schatz. Thank you. I am actually quite excited2846about the prospect that in 20 years people are going to say,2847remember when you had to wait on the phone to talk to Kaiser2848Permanente or the VA?2849    So I just--maybe Mr. Altman and Mr. Smith, I want you to--2850you know, a buddy of mine used to say, paint a picture and2851paint me in it.2852    OK. For the government actually delivering services I want2853you to describe what an AI agent or AI can do to kind of reduce2854those pain points that we accept as a fact of life in2855interacting with the government.2856    It seems to me so much of what makes us irritated with the2857government is the lack of sorting data that exists somewhere2858but we cannot get access to it.2859    So just very quickly, you have 15 seconds each for some2860cheerleading.2861    Mr. Altman. I can imagine a future where the U.S.2862Government offers a AI-powered service that makes it really2863easy to use all government services to get great health care,2864to get great education.2865    You have this thing in your pocket and if you have any2866medical problem you get an answer if you need to, you know,2867like, appeal something on some process you are having with the2868government or file your taxes or whatever. You just do it2869instantly. You have an agent in your pocket fully integrated2870with the U.S. Government and life is easy.2871    Senator Schatz. Anything to add?2872    Mr. Smith. Remember when you had to stand in line to renew2873your driver's license? Remember when you did not know how to2874report a pothole that needed to be repaired on your street?2875    Remember when you had a fender bender in a car and you had2876to fill out all these forms and talk to all these people to get2877insurance coverage?2878    Now you can do it all with one AI system. You can use your2879phone and, by the way, you can do this today in Abu Dhabi. We2880need to bring it to America.2881    Senator Schatz. Thank you.2882    Senator Young [presiding]. Senator Budd.28832884                  STATEMENT OF HON. TED BUDD,2885                U.S. SENATOR FROM NORTH CAROLINA28862887    Senator Budd. Thank you, Chairman.2888    Again, thank you all for being here. I have enjoyed various2889conversations with each of you.2890    The ability for the U.S. to deploy new energy generation2891capacity and upgrade its grid it is in many ways the key to the2892race against China.2893    Energy is how we can win and it is also how we can lose.2894Permitting in this country takes too long. China's command and2895control system means that they will not fail to deploy the2896energy needed to achieve the scale necessary to develop the2897most advanced models which will drive all the benefit of AI.2898    So I am glad to be working with Senator Lummis on the FREE2899Act, which would set up a permit by rule structure which would2900let large projects meet comprehensive standards at the front2901end instead of dragged out on a case by case process.2902    We all want to protect the environment and we all want to2903maintain U.S. economic and technological leadership.2904    So, Mr. Intrator, what has CoreWeave's experience been in2905contracting power and are you concerned that the current2906permitting system can make it harder for the U.S. to achieve2907capital investment in the scale needed to win this AI race?2908    Mr. Intrator. So, as you said, access to power, access to2909scale power is certainly one of the keys to our ability to win2910this race. There are others but it is one that I spend a lot of2911time thinking about.2912    I separated the comment into access to power and access to2913scale power because I do think that we are moving toward a2914period of this race where the size, the magnitude of the2915infrastructure that is being required to move our artificial2916intelligence--the labs that are building it, the companies that2917are building it--forward at the velocity that is necessary is2918going to be a specific challenge that really requires a lot of2919thought.2920    We have a huge part of our organization focused on not just2921getting access to power but getting access to the size and2922scale of power that is going to be able to build the2923infrastructure, you know, at the scale of Abilene or close to2924it in order to, you know, allow this to move forward.2925    It is tough, right, and it will get harder as we move2926through time because the existing infrastructure that does have2927opportunities it has some level of elasticity, is going to be2928consumed, and once that is consumed you are going to get down2929to kind of a first principle how do we get power online now,2930and that is really going to be challenging within the2931regulatory environment as it currently is configured.2932    Senator Budd. Thank you.2933    Mr. Smith, a similar question. How is Microsoft trying to2934secure power for its data centers? I mean, we read about that2935in the news recently but what does Federal policy need to focus2936on to make sure that we do not lose this race because we cannot2937get enough energy?2938    Mr. Smith. Well, we invest to bring more electricity2939generation onto the grid and then to bring it through the grid2940to our data centers. We probably have more permitting2941applications in more countries than quite possibly any company2942on the planet.2943    Last time I looked at it, it was 872 applications in more2944than 40 countries. The number-one challenge in the United2945States when it comes to permitting, interestingly enough, is2946not local. It is not state. It is the Federal wetlands permit2947that is administered by the Army Corps of Engineers.2948    We can typically get our local and state permits done in2949about six to 9 months. The national--the wetlands permit is2950taking often 18 to 24 months.2951    Both the outgoing Biden administration and the incoming2952Trump administration have focused on this, but if we could just2953solve that we could accelerate a lot here in this country.2954    Senator Budd. Very helpful. Thank you.2955    Mr. Altman, much has been made about the Chinese open2956source models like DeepSeek. We spoke about that a month or two2957ago.2958    A concern that I have is that accessible Chinese models2959promoted by the Chinese Communist Party might be an attractive2960option for AI application developers to build on top of,2961particularly in developing world economies.2962    So how important is U.S. leadership in either open source2963or closed AI models?2964    Mr. Altman. I think it is quite important to lead in both.2965We realize that OpenAI can do more to help here so we are going2966to release an open source model that we believe will be the2967leading model this summer because we want people to build on2968the U.S. stack in terms of closed source models.2969    A lot of the world uses our technology and the technology2970of our colleagues. We think we are in good shape there.2971    Senator Budd. So how could Federal policy further help2972encourage the AI ecosystem to be developed right here in the2973U.S.?2974    Mr. Altman. Well, you touched on a great point with energy.2975I think it is hard to overstate how important energy is to the2976future here. You know, eventually chips, network gear, that2977will be made by robots and we will make that very efficient and2978we will make that cheaper and cheaper.2979    But an electron is an electron. Eventually, the cost of2980intelligence, the cost of AI, will converge to the cost of2981energy and it will be how much you can have. The abundance of2982it will be limited by the abundance of energy.2983    So in terms of long-term strategic investments for the U.S.2984to make I cannot think of anything more important than energy.2985You know, chips and all the other infrastructure also but2986energy is where this--I think this ends up.2987    Senator Budd. Thank you. Chairman?2988    Senator Young. Senator Kim.29892990                  STATEMENT OF HON. ANDY KIM,2991                  U.S. SENATOR FROM NEW JERSEY29922993    Senator Kim. Thank you.2994    Mr. Smith, I think I would like to start with you because I2995thought your point about what exactly is the race, right--you2996know, we keep talking about the race, and you framed it in a2997particular way saying that it is about adoption in the rest of2998the world, the 78 percent.2999    I guess I just wanted to ask you to tease that out some3000more in terms of understanding what role we could play in3001Congress, in government, in terms of trying to accelerate and3002champion that AI adoption internationally?3003    Mr. Smith. I think there is two things. The first is it3004just shines light on the importance of getting it right for3005export controls, which is the AI diffusion rule that is being3006discussed right now, and I think what it shows is we want to3007have, I believe, as a country the kinds of national security3008controls that ensure that, say, chips do not get diverted to3009China or get accessed by the wrong users, say, in China for the3010wrong reasons.3011    And that is something that people have drafted in the3012Department of Commerce. At the same time, we need, I believe,3013to say get rid of the quantitative caps that were created for3014all of these tier two countries because what they did was send3015a message to 120 nations that they could not necessarily count3016on us to provide the AI they want and need.3017    And just think about it. I mean, if this is a critical part3018of your country's infrastructure how can you make a bet on3019suppliers if you are not confident that they will be able to3020fulfill your needs?3021    So I think you in Congress and the Senate can help the3022White House and the Department of Commerce get this right.3023    Senator Kim. Mr. Altman, I wanted your thoughts on this. Is3024that the right framing of the race? Is it about the adoption3025internationally in terms of other countries? I guess I am3026trying to think through it.3027    Like, part of what you just said in your previous response3028was that we want other nations to be able to build upon the3029U.S. AI stack. Is that the right framework? Is that what we are3030thinking about?3031    Or is it more about the consumer? Is it more about getting3032the rest of the world and the 78 percent of the population to3033adopt AI applications that are U.S.? Or is it interrelated?3034    Mr. Altman. I think it is heavily interrelated. To me, the3035stack is, you know, from the chips at the bottom to the3036applications on the top, and we want the whole world on the3037U.S. stack. We want them to use U.S. chips. We want them to use3038services like ChatGPT.3039    Senator Kim. Does having other nations building on the3040infrastructure component of the stack--does that more or less3041than guarantee or at least have a high likelihood that then the3042consumers in that country will be using our products and3043applications? Is that the sort of theory of the case?3044    Mr. Altman. It probably does make it marginally more3045likely. But I also think the--if someone is using a stack that3046we do not trust to train models, like, who knows what it is3047going to do?3048    Who knows what sort of back doors would be possible? Who3049knows what sort of, you know, data corruption issues could be3050possible?3051    I think the AI stack is increasingly going to be a jointly3052designed system from the chip all the way up to the end3053consumer product and, you know, lots of stuff in between.3054    I think separating that will not work that well in practice3055and we should not want to. Like, again, I think this point--3056this is a very critical point that the leverage and the power3057the U.S. gets from having iPhones be the mobile device people3058most want and, you know, Google being the search engine that3059people most want around the world is huge.3060    We talk maybe less about how much people want to use chips3061and other infrastructure developed here but I think it is no3062less important, and we should aim to have the entire U.S. stack3063be adopted by as much of the world as possible.3064    Senator Kim. Yes. I mean, when we are looking at--you know,3065you are talking about our investment into models and building3066of that nature.3067    How are we doing in terms of development of the3068application--the AI tools and applications, though, that are3069trying to embed in people's lives?3070    You know, not necessarily just the overarching models but3071do you feel like we are putting the level of intensity that we3072need to in terms of that type of development?3073    Mr. Altman. ChatGPT is the most adopted AI service in the3074world--not just in the United States but in the world--by a3075quite significant margin.3076    We are very proud that people like it and we need to keep3077pushing on that. I think it is important for all the reasons3078you just discussed.3079    There are many other U.S. companies building incredible3080products and services that are also getting globally adopted.3081This is what the U.S. does best.3082    Senator Kim. Dr. Su, I want to just ask one last point to3083you. Over and over again each of you is talking about talent as3084this incredible power but also could be a bottleneck to us.3085    How are we doing when it comes to development of talent in3086this country? If you were to give us a grade what would you3087grade us at in terms of our development right now?3088    Ms. Su. Thank you, Senator, for the question. Look, I think3089the smartest engineers are in the United States. We have a3090great base of talent.3091    But what I will say is we need more. We need more hardware3092developers, software developers, application developers.3093    Senator Kim. How wide is that delta? If we are talking3094about this as a race, as you did, you know, is that a space3095where we have a larger amount of delta or is that a place where3096it is closing rapidly, too?3097    Ms. Su. Well, I think we do have a very talented overall3098talent base but we also have the desire to have the best and3099that includes not only, you know, U.S. nationals but also3100having the best international students.3101    Senator Kim. Drawing the talent from----3102    Ms. Su. That is right. I think high-skilled immigration is3103one of those areas where we want the best people in the world3104to be doing their work in the United States.3105    And, Senator, if I can just add something to your previous3106point about the cycle and what race we are trying to win.3107    You know, technology is one of those things where you can3108have a very vicious positive cycle. So, in other words, when we3109lead and more people adopt that means more developers that make3110our technology better.3111    That increases our lead. So that is what we want is to have3112our leadership just increase over time.3113    Senator Young. Senator Schmitt.31143115                STATEMENT OF HON. ERIC SCHMITT,3116                   U.S. SENATOR FROM MISSOURI31173118    Senator Schmitt. Thank you, Mr. Chairman.3119    Mr. Altman, I will start with you. I really enjoyed and was3120inspired by your story with the light on in the home you grew3121up in in St. Louis and you talked about the spirit of3122innovation. That is the Spirit of St. Louis.3123    As a fellow St. Louis native that is a good story to hear,3124and we just look forward to more investment in St. Louis from3125your company. That would be great, too. So I will put a plug in3126for that.3127    But I do want to ask you specifically, there is a lot made3128of sort of the comparison between the United States and the3129regulatory environment and what exists in Europe.3130    What specifically--and I will open this up too--what3131specifically has gone wrong in Europe that we can draw some3132conclusions from?3133    Mr. Altman. First of all, we would love to figure out how3134to invest more in St. Louis. I would love an excuse to get to3135go home more often.3136    I will point out one example that I think is just very3137painful to users. When we launch a new feature or a major new3138model we have what is now considered a little bit of like an in3139joke where we say we have this great new thing not available in3140the EU and a handful of other countries because they have this3141long process before a model can go out.3142    And there will be, I believe, great models and services3143that are quite safe and robust that we will be unable to offer3144in other regulatory regimes, and if you are trying to be3145competitive in this new world and if you are consistently some3146number of months behind what other people in other countries3147get access to, that is an example that is extremely painful to3148users.3149    Senator Schmitt. And you mentioned sort of your observation3150that the AI stack may make it more vertically integrated. So3151how does that work then?3152    Because right now the best estimates, I suppose, right, is3153that--I do not know, with China is 2 months to 6 months behind3154maybe on large language models. Hopefully, some of the advances3155we are seeing in the U.S. maybe there is a degree of3156separation. It is hard to know exactly, right, with DeepSeek.3157    But then you get down to the chips and that advantage is3158more like a couple of years probably, something like that.3159    So if that is where we are headed does that increase the3160U.S.' advantage, in your view, or does that sort of allow China3161to catch up quicker as we get more vertically integrated?3162    Mr. Altman. I think there are a lot of things that can3163increase U.S. leadership. But we touched on this earlier--I3164think it is so important. There will be great chips made around3165the world. There will be great models trained around the world.3166    If the United States companies can win on products and3167the--sort of all of the positive feedback loops that come from3168how you can improve this once, you know, real users are using3169your products in their daily lives for their hardest tasks,3170that is something special that is not so easy to catch up with3171just by doing good chips and good models.3172    So making sure that the U.S. can win at the product level3173here. Obviously, I am, like, talking my book a little bit, but3174I really do believe it--is quite important, and that is in3175addition to all of the chips algorithm or the infrastructure3176algorithms and data. I think this is a new area where the U.S.3177is really winning and has a very strong compounding effect.3178    Senator Schmitt. Mr. Intrator--did I pronounce that3179correctly, by the way?3180    Mr. Intrator. Yes.3181    Senator Schmitt. OK, thank you.3182    I want to turn a little bit, sort of staying on this3183regulatory environment, one of the things I think that is most3184concerning that is coming out of Europe is this sort of3185censorship regime that exists not just online but in real life.3186    But, certainly, it is happening online. I mean, people are3187being arrested for things that they say online, and one of the3188concerns I have with AI, I suppose, is that if we end up with a3189place where it is somehow policing, quote/unquote,3190``misinformation'' and, you know, I think even in NIST's most3191recent voluntary standards one of the risks to be on the3192lookout for was the spread of misinformation.3193    So the point of the question is how do we make sure that--I3194think part of what is going wrong in Europe is it is a sort of3195a--it is funneling information and, in my view, whether I agree3196with the point of view or not it ought to be out there. People3197can make their own decisions. You combat speech you do not3198agree with not by censoring it but by--with more speech.3199    What are some lessons to be learned there and make sure3200that does not happen here?3201    Mr. Intrator. So Europe is moving forward with its3202regulatory regime in a European way, and from our seat where we3203have to make these enormous capital investments one of the3204things about the approach that Europe is taking that we are3205deeply concerned about every day is the balkanization to use--3206of how they go about allowing information to flow and how they3207go about regulating it, how they go about with each component3208of their union having its own set of rules, which will be3209tremendously challenging in Europe as time goes on because it3210is really hard to make the magnitude of investments that we--3211where we are----3212    Senator Schmitt. Beyond that, though, jurisdictionally I am3213talking about content now.3214    Mr. Intrator. So we are not--the role of our company is3215really kind of below that, you know--and, you know, Sam and,3216you know, Microsoft you are going to get a lot more attention3217paid to the content level because of the role that they play in3218the stack. It is not really where we are primarily focused. We3219are really focused on the investment side of it.3220    Senator Schmitt. Yes. If any of you would like to--Sam, if3221you--or, Mr. Altman, if you would like to respond to that I3222would like to get some answer.3223    Mr. Altman. I think--well, first of all, I strongly agree3224that people getting, you know, like, put in jail for stuff they3225say online is very--not American and not what we should be3226doing.3227    AI is quite different than social media, at least in its3228current evolution. People are using these tools in this sort of3229one on one way instead of this massive thing online.3230    So I think it is easy to make too many analogies but it is3231a little bit dangerous to try to talk about AI and the things3232we are going to face here in the same way that we did for3233social media. But our stance is that we need to give adult3234users a lot of freedom to use AI in the way that they want to3235use it and to trust them to be responsible with the tool.3236    And I know there is increasing pressure in other places3237around the world and some in the U.S. to not do that but I3238think this is, like--this is a tool and we need to make it a3239powerful and capable tool.3240    We will, of course, put some guardrails and very wide3241bounds but I think we need to give a lot of freedom here.3242    Senator Schmitt. Yes. I am out of time but there is a lot3243more questions there that we will follow up with.3244    Thank you, Madam Chair.3245    Senator Cantwell [presiding]. Thank you. Thank you.3246    Senator Hickenlooper.32473248             STATEMENT OF HON. JOHN HICKENLOOPER,3249                   U.S. SENATOR FROM COLORADO32503251    Senator Hickenlooper. I appreciate that line of3252questioning. I was ready for you to continue as well. I could3253have given you a minute or two.3254    Mr. Smith, Microsoft has a long and deep history in3255transforming workplaces all over the world through software,3256from Windows operating system to its office applications like3257PowerPoint, Excel, and now the AI-powered Copilot application.3258    In software development, life cycles seem to be becoming3259increasingly shorter, updates becoming more frequent.3260    What are the internal processes that Microsoft follows to3261evaluate Copilot's accuracy and performance before it was3262released and what kind of independent review teams other than3263Microsoft's own product developers are involved in that? Who do3264you bring in to help with that?3265    Mr. Smith. Well, first of all, since most of what we are3266talking about here when you are talking about our Copilot's3267start with models that are developed at OpenAI, I would say3268OpenAI has its internal process.3269    There is then a joint--what is called the DSB, a Deployment3270Safety Board, where we decide together whether something is3271safe to deploy, as the name implies.3272    We then at the applications level have our own internal3273Deployment Safety Board. We have a variety of engineering tools3274that we use to assess these features. We test these features.3275    We have red teams, meaning sort of competing teams that3276often go to work to sort of attack the features, and then3277ultimately the product is released when those tests are3278completed and the results are satisfactory.3279    Senator Hickenlooper. I like that. Well, let me go over to3280Mr. Altman. Obviously, you all have a natural incentive to3281ensure that the products are high quality and safe.3282    But the field is so competitive and, you know, in applied3283research and with rigorous testing these constant improvements3284really are fundamental steps to the performance of a model.3285    So risk assessments are that key tool, and I am a big3286believer in evidence-based technical standards. I have been3287accused of being the only real scientist who has published3288peer-reviewed papers in the Senate.3289    So, Mr. Altman, do you believe that under appropriate3290circumstances independent evaluations based on standards3291performed by qualified evaluators and done voluntarily could3292help validate the testing that you are performing internally3293and in conjunction with peer companies?3294    Mr. Altman. Thank you, Senator, and I think it is awesome3295that you are--have published peer-reviewed papers and would3296love to see more of that.3297    Senator Hickenlooper. I was--on the Maslow's triangle of3298science I was near the bottom. I was a geologist. So that is3299not high up in that----3300    Mr. Altman. Geology is great.3301    Yes, I think what you say is very important. It is an3302important part of our process today. External testing helps us3303find things that we may have missed internally and--we are very3304proud of our safety record on the whole, not that we--you know,3305we have not been perfect and we are continuing to learn new3306things, but I think we do have a process that is leading toward3307models that the public generally thinks are safe and robust to3308use, and we have developed a lot of techniques to be able to3309continue to deliver that.3310    But external testers and red teamers are a critical part of3311that process and I think they have helped us find many things3312in the models to improve.3313    Senator Hickenlooper. Mr. Smith, would you add anything to3314that?3315    Mr. Smith. No.3316    Senator Hickenlooper. OK. Got it. Someone giving testimony3317who does not have something to add--it is a moment of3318scientific reflection.3319    Dr. Su, the bipartisan CHIPS and Science Act, an historic3320effort to try and maintain U.S. leadership in emerging3321technologies like semiconductors but others as well, as the3322technology arms race continues globally, and you were talking3323about this, AMD plays a key role in delivering state-of-the-art3324designs, the best for the new chips that are going to power our3325electronics and the devices that are going to allow AI to3326become global.3327    As scientists work around the clock to develop new3328breakthroughs and to try to increase and improve performance3329but at the same time shorten R&D timelines, what do you see as3330the next frontier of chip technology in terms of energy3331efficiency and how can--and that is not just based on the3332Chinese competitors but how can we work together to improve3333direct to chip cooling for high-performance computing?3334    Ms. Su. Well, thank you for the question, Senator.3335    I would say, look, there is a tremendous amount of3336innovation that is going on in the semiconductor sector today.3337The CHIPS and Science Act was certainly helpful in raising the3338profile of chips in the United States.3339    Relative to, you know, what are we doing to go faster and3340build better and more power efficient chips, frankly, we are3341using AI extensively through our chip development cycles and it3342does allow us to augment what are typically very long cycles,3343many years--you know, several years for us to develop chips.3344    We can shorten the time and also improve the efficiency,3345and there are lots and lots of great new technologies in terms3346of cooling technologies that are super important for us to3347build the large-scale systems that we talked about earlier3348today. So thank you for the question.3349    Senator Hickenlooper. All right. I am out of time. I will3350yield back to the Chair. Thank you all.3351    The Chairman [presiding]. And, Senator Hickenlooper, I will3352say as a Texan whose parents were in the oil and gas business I3353think geologists are awesome.3354    Senator Hickenlooper. We have a consensus.3355    The Chairman. Senator Curtis.33563357                STATEMENT OF HON. JOHN CURTIS,3358                     U.S. SENATOR FROM UTAH33593360    Senator Curtis. Thank you, Mr. Chairman. It is a delight to3361be here.3362    Mr. Altman, you started kind of a one-upmanship on3363computers and I will just tell you in 1985 the month you were3364born I was attending a class at Brigham Young University and3365carried in a laptop and was almost kicked out.3366    Mr. Altman. What laptop?3367    Senator Curtis. It was a TRS 80----3368    Mr. Altman. Oh, awesome.3369    Senator Curtis.--made by Radio Shack. I upgraded the memory3370from 40K to 80K. Ran on four AA batteries and------3371    Mr. Altman. That is incredible.3372    Senator Curtis. Yes. So I am very envious of your3373generation. Let me start with you, if I would.3374    I think, you know, Utah would aspire to lead out with data3375centers and advanced technologies. Could you just address for3376states and Utah specifically what it is that makes them3377attractive to projects like Stargate?3378    Mr. Altman. Yes, and I know that we are having productive3379discussions about some potential sites in Utah. Power cooling,3380fast permitting process, labor force that can build these3381things--the electricians, the construction workers, the entire3382stack.3383    A state that wants to, like, partner with us to move3384quickly. Texas really has been unbelievable on this. I think3385that would be a good thing for other states to study but we3386would be excited to try to figure something out.3387    Senator Curtis. Thank you. I think I could speak for our3388state leaders. We would be excited as well. But as you know,3389this also brings challenges and one of those challenges are the3390demands for energy, and what are your thoughts on how we3391protect rate payers and kind of put a little bit of a firewall3392between them?3393    Mr. Altman. I mean, I think the best way is just much more3394supply, more generation. You know, like, I think if you make it3395easy to reasonably profitably create a lot of additional3396generative capacity the market will do that.3397    That will not only not drive up rates because of the AI3398workload--hopefully it will drive it down for everything.3399    And we have talked a lot about the importance of energy to3400AI. Energy is just really important to quality of life. One of3401the things that seems to me the most consistent throughout3402history is every time the cost of energy falls the quality of3403life goes up, and so doing a lot to make energy cheaper in the3404short term--I think this probably looks like more natural gas,3405although there are some applications where I think solar can3406really help in the medium term.3407    I hope it is advanced nuclear fission and fusion. More3408energy is important well beyond AI. You know, in some sense we3409have these dual revolutions going of AI and energy, the ability3410to have new ideas and the ability to get them done, to make3411them happen in the physical world where we all live. Like,3412these are kind of the limiting reagents of prosperity and let3413us have a lot more.3414    Senator Curtis. Thank you.3415    Mr. Smith, we have talked about how significant power was--3416is to the success here. What role do you think Microsoft and3417other tech leaders have in developing energy and particularly3418the right type of energy?3419    Mr. Smith. I think we have a tremendous responsibility to3420contribute to the solution and I think Sam helped with his3421list.3422    I would highlight two things, and I just would, I guess,3423illustrate it with what we do everywhere but most recently with3424a major site in southeastern Wisconsin. We went from zero,3425basically, to becoming the largest industrial user of3426electricity in the state--roughly, 400 megawatts.3427    And so we worked with the local utility. We made the3428investment to help and really enable them to expand their3429electricity generation.3430    Now, that electricity then needed to be delivered from3431their power plant through the grid to our data center. We went3432to the Public Utilities Commission and we proposed a rate3433increase on ourselves because we thought it was important that3434we pay for that improvement to the grid so that the neighbors,3435so to speak, would not have to.3436    And I think what it really illustrates is the collaborative3437partnerships that are needed to provide the capital, to do the3438construction, to improve the grid, and to be, I think, very3439sensitive to the community as a whole.3440    Senator Curtis. Thank you.3441    Mr. Altman, let me come back to you. I was a small business3442owner. I have a special spot in my heart for small business3443owners.3444    Can we talk a little bit about ChatGPT and how that might3445assist small business owners? And let me paint a little broader3446picture. We have heard a lot about other tools that are,3447perhaps, out of favor, particularly with the U.S. Government,3448that are very helpful for small businesses.3449    But I do not know if small businesses are fully3450understanding the platform that you have and how they might use3451it for marketing, for data research, and ways to help their3452small business be successful.3453    Mr. Altman. One of--there were all these moments as ChatGPT3454was beginning to take off where we would be, like, oh, we may3455have, like, a hit on our hands.3456    There is, like, that is--someone is using it for this and3457this and that, you know, strangers talking about it. You see3458someone using it in a coffee shop.3459    But one of the ones that really sticks out for me is pretty3460quickly after ChatGPT launched, like, in the first six months,3461say, I was in an Uber and the driver was making conversation.3462    He is, like, have you heard of this thing called ChatGPT?3463It is amazing. And I was, like, yes. Like, what do you think3464about it?3465    And he was using it to run basically his entire small3466business. He was, like, I had--he ran a Laundromat and he was,3467like, I had all these problems, you know, like, could not find3468good people to write my ads, could not get, like, legal3469documents reviewed, could not, like, answer customer support e-3470mails.3471    And he was, like, a mega early adopter but he was one of3472these people that was using AI to, like, make a small business3473work and that was--we talked about that story a lot at the time3474but it is nice to reflect on it again now.3475    We have now heard that at scale from a lot of people, but3476that was one of those moments early on we were, like, oh, this3477is maybe going to work.3478    Senator Curtis. So and I am out of time but just to mark3479this is more than just something that helps proofread e-mails,3480right? And you do not need to comment because I am out of time,3481but I think we would all agree with that.3482    Mr. Altman. It is.3483    Senator Curtis. And look forward to seeing these3484applications move forward.3485    Mr. Chairman, I yield my time.3486    The Chairman. Senator Duckworth.34873488              STATEMENT OF HON. TAMMY DUCKWORTH,3489                   U.S. SENATOR FROM ILLINOIS34903491    Senator Duckworth. Thank you, Mr. Chairman. I thank you,3492the panel, for all of you being here today.3493    I want to begin by talking about the importance of3494partnerships between the private sector and our National3495Laboratories in maintaining United States leadership in AI.3496    Illinois is the proud home of two crown jewels of the3497National Laboratories, Fermilab, America's premier particle3498physics and accelerator laboratory, and Argonne National3499Laboratory, home to the Aurora supercomputer that will3500accelerate breakthroughs in AI, cancer research, and3501fundamental physics.3502    There is nothing more important than sustaining and3503amplifying investments in our Nation's incredible network of3504National Labs.3505    Yet, Donald Trump and Elon Musk, with the support of some3506Republicans in Congress, are plotting to take a chainsaw to the3507vital research initiatives being carried out across our3508country.3509    This is a self-sabotaging attack, plain and simple, and if3510allowed to proceed Trump and Musk will inflict lasting harm on3511our innovative capabilities and capacity that our enemies could3512only dream of achieving.3513    Does anyone truly have confidence that had DOGE been around3514decades ago they would not have cut the project that created3515the Internet as an example of wasteful publicly funded research3516and development?3517    So my question to any member of the panel is the following.3518Can you explain the importance of the National Labs system to3519maintaining our research edge and discuss any partnerships you3520have established or are currently pursuing, especially those3521threatened by massive cuts to the National Labs' research?3522    Mr. Altman. We partner with the National Labs so maybe I3523could take a first cut of this.3524    Senator Duckworth. Please.3525    Mr. Altman. Also, Senator, I would love to get to visit3526Fermilab someday. That would be, like, unreal.3527    Senator Duckworth. That was my next question. You are3528welcome.3529    Mr. Altman. That would be a real life highlight. That would3530be very cool.3531    There are many wonderful things that AI is going to do for3532the world but the one that I am personally most excited about3533is the impact AI will have on scientific discovery. I believe3534that new scientific discovery is the most important input to3535the world getting better and people's quality of life getting3536better over time.3537    It is hard to overstate where we would be if--where we are3538because of scientific advancement and where we would be without3539it. So we are thrilled to get to partner with the National Labs3540on this.3541    I think science has not been as efficient as it can be, and3542we are also thrilled to hear from scientists that they are, you3543know, multiples more effective than they used to be and I think3544that AI tools will mean we can accomplish at some point a3545decade worth of scientific progress in a year for the same cost3546or even less.3547    This will be one of the most important contributions, in my3548opinion, that AI makes to the world. And it is no longer3549theoretical. Like, the National Labs are a great example.3550    It is the only partnership where we have given a copy of3551our model weights to another organization. It is a very deep3552and important partnership to us and I expect that that will3553really bear fruit.3554    Senator Duckworth. Thank you. Anybody else on the panel?3555    Mr. Smith. Yes, I think you highlight a very important3556issue. This country has 17 National Labs administered by the3557Department of Energy and about 85 to 90 research universities,3558and together they are the fabric of much of scientific3559discovery and have been since the Manhattan Project in World3560War II.3561    We in the tech sector, we at Microsoft, work with most,3562almost all of them, and there is a particular cycle of3563innovation that the United States has mastered. You have3564curiosity-driven research in these institutions and then the3565advances move out of those institutions into startups and into3566larger companies.3567    And what I always find interesting, as I meet with3568officials around the world they have studied this. They seek to3569emulate it, and I always worry that in the United States we run3570the risk of taking it for granted.3571    We should never take this for granted. It is the foundation3572for the country's technological leadership.3573    Senator Duckworth. Very much so.3574    Dr. Su.3575    Ms. Su. I just wanted to add to that. We are also very3576large supporters of the public-private partnerships with the3577National Labs.3578    I think the National Labs have, you know, in a way always3579tried to look ahead of the curve and, you know, that is a great3580place for us to invest.3581    We think they are a key piece. We have partnered with all3582of the National Labs as well, you know, over the last decade3583and that continues to be a place where I think there can be3584significant public-private partnership.3585    Senator Duckworth. Thank you. Mr. Intrator.3586    Mr. Intrator. I just think it would be really interesting3587to come to these AI factories and to walk or travel through3588these institutions and identify all the different pieces of the3589science that leads back and was ultimately driven and founded3590on something that came out of those institutions. It is3591amazing, actually.3592    Senator Duckworth. Thank you. And would any of the3593remaining three of you like to come to a lab in Illinois,3594either Fermi or Argonne? I will give you personal tours.3595    [Laughter.]3596    Senator Duckworth. All right. All four of you. It is done.3597    Thank you, Mr. Chairman.3598    The Chairman. Thank you.3599    Senator Young.36003601                 STATEMENT OF HON. TODD YOUNG,3602                   U.S. SENATOR FROM INDIANA36033604    Senator Young. Thank you, Mr. Chairman, for holding this3605important hearing on winning the AI race. It is good to see our3606panelists here.3607    One of the things that I like to underscore whenever I talk3608about this issue is we are not just discussing a race to create3609jobs, not just discussing a race to figure out how to eke out3610more growth from our economy, although that is important.3611    Not just trying to identify how humans can flourish more,3612especially Americans, through application of AI solutions to3613our daily lives in various ways.3614    But this is an issue of national and economic security. I3615want folks at home to get that. I know all our panelists are3616highly conversant and knowledgeable about that.3617    In my discussions with you and many others I have heard we3618need to work with like-minded partners and allies to win this3619race, and it is only going to be done collectively.3620    I have heard here today from a number of you that this race3621is in part about getting market share, diffusion of our AI3622models and solutions into other countries.3623    It is through that means for me to, perhaps, elaborate on3624your thoughts that we can see that our own values are advanced.3625    These models presumably they will be embedded with our3626values related to privacy and transparency and property rights3627and freedom of speech and religion, not the values of the3628Chinese Communist Party on each of those various fronts.3629    And then if we can establish digital trade rules, digital3630cross-border agreements on digital trade with these other3631countries, we could conceivably erect higher barriers to entry3632for models that do not come embedded with our standards, models3633of, say, the Chinese Communist Party has given sanction to.3634    So there is a geopolitical national security overlay to3635this entire conversation, which is why I think the Chairman's3636emphasis on not overly constraining innovation or deployment is3637very important.3638    But it is also why I think it is important that we be3639thinking about how to work with other countries in their3640standards development.3641    And so that is that is where I want to begin asking3642questions. I will start with Mr. Smith.3643    If the United States does not adopt some standards through3644some entity, whether it is NIST or another Federal entity or3645federally sanctioned entity, then won't other nations go ahead3646and feel the need to adopt their standards without any3647consultation with the United States?3648    Mr. Smith. I think it is a really important point you make3649and it is the lesson from the evolution of privacy law. The3650United States did not adopt a national privacy law.3651    Europe did twice, and most American companies of any size3652today apply across the United States work that complies with3653European privacy law. It is just more efficient.3654    So I think the United States needs to be in the game3655internationally to influence the rest of the world, and you3656cannot be in the game if you do nothing. You must do something.3657    So you take Senator Cruz's idea--a lightweight approach----3658    Senator Young. Yes.3659    Mr. Smith.--and then you build support around it.3660    Senator Young. So just to unpack that--and I will stick3661with Mr. Smith with apologies to everyone else because my time3662is limited--would it be easier to shape the standards of other3663large economy countries that share most of our values if we3664already have a set of standards adopted?3665    Mr. Smith. Generally, yes. I think we always have to be3666careful because if you go too soon you go before the standards3667have really come together. But you have got to have some kind3668of model that you can show the rest of the world and win3669support for.3670    Senator Young. And then presumably standards could be3671harmonized, right? They are not set in--and chiseled onto a3672tablet, so to speak, right?3673    Mr. Smith. That is indispensable. I mean, if our technology3674is going to go around the world we need a set of laws or3675regulations that, in effect, create that basis for reciprocity3676and interoperability.3677    Senator Young. OK. I only have 25 seconds left. Are there3678any violent objections to Mr. Smith's position? Because that3679seems eminently reasonable to me.3680    Seems consistent with the light touch approach but it also3681shows a certain sense of urgency that the United States needs3682to act.3683    The last thing I will say in my remaining 10 seconds is3684that I am planning on introducing legislation today called the3685AI Public Awareness and Education Campaign Act with several of3686my colleagues and our aim is to have a whole of government3687approach to foster greater awareness of AI literacy and grow3688STEM opportunities to create the next generation of our3689workforce, and looking forward to moving that forward.3690    So it will be available for public review, critique, even3691accolades and, Mr. Chairman, I yield back.3692    The Chairman. Thank you.3693    Senator Blunt Rochester,36943695            STATEMENT OF HON. LISA BLUNT ROCHESTER,3696                   U.S. SENATOR FROM DELAWARE36973698    Senator Blunt Rochester: Thank you, Chairman Cruz, and3699thank you so much to the witnesses. This is such an important3700hearing. Five minutes will not suffice for me. I will be3701submitting some questions for the record.3702    I notice that for Mr. Altman and Mr. Smith when the3703question of paint me a picture of the future came up there3704was--you actually leaned up in your chair. There was a level of3705excitement, and that is how I am about the future.3706    When I came into the House of Representatives in 2017 I3707started a Future of Work bipartisan caucus because I had a3708concern that, number one, there were certain groups of people3709that were going to be left behind but there--also as a country3710that we could be left behind.3711    And I started--I had an event where we had everyone walk3712into the room and use a word cloud and tell me what you think3713of when you hear the future of work. The biggest word coming in3714the door was fear. The biggest word walking out the door was3715opportunity.3716    And so, to me, this conversation is so vital to think about3717the opportunities but also making sure that we are watching out3718for ethics, watching out for scams, watching out that3719technology does not take over the human.3720    And so I am just grateful for the conversation and, Mr.3721Altman, I listened to an interview about--that you gave with3722Lester Holt maybe a year or so ago and you talked in that3723interview about how OpenAI--it was not initially even about3724making a product.3725    It was not about the money. And so I know you are3726incorporated in Delaware and I understand you have been working3727with our attorney general during the previously proposed3728legislation to transition to a for profit--not legislation but3729to transition to for profit--and this Monday, OpenAI decided to3730apply to become a public benefit corporation instead and to3731have the PBC govern your nonprofit arm.3732    What went into this decision and what considerations3733influenced the timing of the organizational change?3734    Mr. Altman. So we never--thank you for the question,3735Senator, and the chance to explain this. It is a complicated3736thing that I think has gotten misrepresented. So this is a3737wonderful forum to talk about it.3738    We never planned to have the nonprofit convert into3739anything. The nonprofit was always going to be the nonprofit,3740and we also planned for a PBC from the very beginning.3741    There were a bunch of other considerations about is it the3742PBC board that would control the nonprofit somehow or, you3743know, how our capital structure was going to work that there3744was a lot of speculation on most of it, inaccurate in the3745press.3746    But our plan has always been to have a robust nonprofit. We3747hope our nonprofit will be one of the best, maybe someday the3748best resourced nonprofit in the world, and a PBC with the same3749mission that would make it possible for us to raise the capital3750needed to deliver these tools and services at the quality level3751and availability level that people want to use them at but3752still stick to our mission, which we have been proud over the3753last almost decade of our progress toward.3754    So we had a lot of productive conversations with a lot of3755stakeholders and a lot of lawyers and a lot of regulators about3756the best way to do this.3757    It took longer than we thought it was going to. You know, I3758would have guessed that we would have been talking about this3759last year. But now we have a proposal that people seem pretty3760excited about and we are trying to now advance.3761    Senator Blunt Rochester. And, Dr. Su, your company3762primarily operates in the physical hardware portion of the AI3763stack.3764    I have a bill with Senators Cantwell and Blackburn called3765the ``Promoting Resilient Supply Chains Act'', which authorizes3766the Department of Commerce to strengthen American supply chains3767for critical industries and emerging technologies.3768    Dr. Su and others, semiconductor and chips manufacturing is3769critical to advancing the advancement of AI but we are facing3770these global supply chain constraints.3771    What specific policies--and I know you mentioned policies3772as well for supply chains--would we need to adopt to help3773American companies overcome the supply chain issues and compete3774in international with our rivals?3775    Ms. Su. Thank you, Senator, for the question.3776    There is no question the semiconductor supply chain and3777overall supply chains are really critical for us to win the AI3778race. I think from a semiconductor standpoint the efforts that3779have been made to move manufacturing back to the United States3780have been positive.3781    I think they are a start. There is a lot more that we can3782do, and one of the most important aspects of it is really to3783think about it end to end.3784    There are so many steps to go from beginning to end in a3785semiconductor supply chain including advanced wafers, including3786packaging, including the back ends and system tests.3787    All of those avenues need to have a footprint in the United3788States, and then we have many allies around the world which3789are, you know, very excellent partners as part of the global3790resiliency in the supply chain and we would like to see those3791partnerships continue to flourish.3792    Senator Blunt Rochester. Last question, if I can.3793    Mr. Smith, how do you see the interdependence between the3794AI stack sections creating either vulnerabilities or3795opportunities in the AI supply chain?3796    Mr. Smith. I think they create more opportunities than3797vulnerabilities because it enables companies to do what they do3798best and that we can work together.3799    And the world today has an integrated supply chain for3800anything that you buy. We just do not think about it when we go3801to the grocery store.3802    I think one of the strengths of the tech sector is that we3803have--I will call it a string of pearls, great companies in3804very--in every layer of the stack and we are going to need,3805frankly, more great companies, especially at the applications3806layer, and that it is how we work together.3807    Senator Blunt Rochester. Thank you so much. I am out of3808time but we will be following up with questions for the record3809as well as individually. Thank you, and I yield back.3810    Senator Lummis [presiding]. Mr. Moran.38113812                STATEMENT OF HON. JERRY MORAN,3813                    U.S. SENATOR FROM KANSAS38143815    Senator Moran. Chairman Lummis, thank you very much.3816    Mr. Smith mentioned data privacy, which has been a topic of3817mine for a long time, and we have been unsuccessful in3818legislation being adopted. But I still have the goal of making3819certain that consumers have control over their own data.3820    And I was going to ask you, Mr. Altman, how can we provide3821consumers with more control over how their data is used by AI3822companies while preserving the utility of the AI system? So how3823do you get more privacy and still get the benefits?3824    Mr. Altman. So there is all of the standard privacy3825controls that companies like ours and others build and should,3826but there is a new area that I would love to flag for your3827consideration, which is people are sharing more information3828with AI systems than I think they have with previous3829generations of technology, and the maximum utility of these3830systems happens when the model can get very personalized to3831you.3832    So this is a wonderful thing and we should find a way to3833enable it, but the fact that these AI systems will get to know3834you over the course of your life so well I think presents a new3835challenge and level of importance for how we think about3836privacy in the world of AI--how we are going to think about3837guaranteeing people privacy when they talk to an AI system3838about whatever is happening in their lives--how we make sure3839that when one system connects to another it shares the3840appropriate information and does not share other information3841and that users are in control of that.3842    I believe this will become one of the most important issues3843with AI in the coming years as people come to integrate this3844technology more into their lives, and I think it is a great3845area for you all to think about and take quite seriously.3846    Senator Moran. We do. We just do not have any success in3847finding the conclusions. But thank you for the encouragement.3848    I chair a Commerce, Justice, Science Appropriations3849Subcommittee that funds the Department of Justice and it plays3850a significant role in cybersecurity of our country.3851    I just came in from a budget hearing with the FBI Director3852Dr. Patel in which we covered cybersecurity threats.3853    AI can--and I think this is true--AI can be used on both3854sides of a cybersecurity attack and it can be used to automate3855phishing, malware creation. But machine learning can also3856increase our ability to detect and respond to cyber threats.3857    What should Congress think about allocation of Federal3858resources for cybersecurity and what should we consider when it3859comes to AI?3860    Mr. Smith. I would say that AI, as you said, is both an3861offensive weapon and a defensive shield when it comes to3862cybersecurity, and as with many other things the front line of3863this the last few years has been in Ukraine because Russia has3864such a sophisticated cyberattack capability.3865    And, you know, what we have found is a company that has3866been involved in supporting Ukraine since literally the moment3867that war began is that AI is a game changer.3868    We have intercepted attacks against Ukraine faster than a3869human could detect them and we block those attacks from taking3870place.3871    So you deploy AI into--call it the front line of the3872products themselves. We have to recognize that it is ultimately3873the people who defend not just countries but companies and3874governments, the chief information security officers, or the3875CISOs.3876    So we have created what is called a cybersecurity copilot3877that basically automates for those individuals much of the3878workflow that takes their time so that they can be more3879effective and efficient.3880    When it comes to Federal appropriations I think that, to3881put it simply, the U.S. Government must remain at the forefront3882of having for itself the cybersecurity capabilities that it3883needs to defend the government and every day--I mean, we are in3884government agencies today during this hearing, you know,3885pushing Chinese out of agencies and the like, and this will3886happen every day of every year from now to probably eternity.3887    So we must keep the U.S. Government well funded in this3888space and I think we also need our intelligence agencies and3889especially the NSA to be well funded so they can remain at the3890forefront when it comes to global leadership in this field.3891    Senator Moran. Thank you for your observations and3892encouragement.3893    My final question--rural areas, a place I come from, often3894lack high-speed broadband, and since many AI tools rely upon3895connectivity I am concerned that many parts of the country and3896many parts of Kansas may not be able to access the benefits3897that AI will bring to business, schools, health care, et3898cetera.3899    What can the Federal Government do to be supportive of3900development and availability of on device or low broadband3901width AI systems that do not rely on constant connectivity?3902    Mr. Altman. I am generally pretty excited about what AI3903will do here because you can offload so much of the processing3904to the cloud and then ship a relatively small amount of data.3905    If you think about, you know, ChatGPT as text comes in3906there is, like, a brain that thinks about it really hard and3907some text comes back, we can support people in low connectivity3908areas quite well with the same quality of service.3909    Separately to that, I think getting great connectivity3910everywhere is important but in the specific area of AI I think3911we can actually address that gap quite well.3912    Senator Moran. That is good to know.3913    Thank you very much.3914    Senator Lummis. Mr. Lujan.39153916               STATEMENT OF HON. BEN RAY LUJAN,3917                  U.S. SENATOR FROM NEW MEXICO39183919    Senator Lujan. Thank you, Madam Chair.3920    And, first, I want to begin by recognizing and thanking Mr.3921Altman and Mr. Smith for your organizations' ongoing3922involvement in the NIST USAI Safety Institute, as well as Dr.3923Su and Mr. Altman for your ongoing partnerships with our3924National Laboratories.3925    Now, Dr. Su and Mr. Altman, can you explain how your3926partnership with the National Labs support scientific research?3927You explained this to a question that was asked by Senator3928Duckworth as well but if you could just touch on that quickly.3929    Mr. Altman. Our latest models, like, 2003 are good at3930scientific reasoning and so scientists are able to use these to3931help them review literature, come up with new ideas, propose3932experiments, analyze data in a way that the previous3933generations of models just could not.3934    We have had the National Labs and other scientists spend3935time with previous models and they say, oh, this is, you know,3936kind of cool. It is interesting. It is not transforming things.3937    These new models are the first time we are hearing from3938scientists at the National Labs and elsewhere that this is a3939legitimate game changer to their research output.3940    Senator Lujan. I appreciate that. Dr. Su?3941    Ms. Su. Yes, I would add the same. I think our partnerships3942with the National Labs have seen just tremendous opportunity.3943We have large-scale compute across the National Labs and the3944ability to really develop new applications that take advantage3945of, let us call it traditional high-performance computing,3946together with the new AI model capability that we just talked3947about is, I think, a great opportunity to substantially move3948forward the ability for scientific discovery.3949    Senator Lujan. To both of you, again, can you explain why3950Federal investment in foundational research and standards3951bodies are crucial to your companies?3952    Mr. Altman. I think standards can help increase the rate of3953innovation, but it is important that the industry figure out3954what they should be first. I think a bad standard can really3955set things back and we have seen many examples of that in3956history.3957    I do think there is a new protocol to discover here at the3958level of importance of HTP. This is just one example. There is3959many other things, too.3960    I believe the industry will figure that out through some3961fits and starts and then I think officially adopting that can3962be helpful.3963    Senator Lujan. Dr. Su.3964    Ms. Su. I believe public-private partnerships really enable3965us to think, let us call it, ahead of the curve. So there are3966lots of things that we do in industry and we do them very, very3967well.3968    However, the beauty of the National Labs and Federal3969research is it does allow let us call it a bit more bluesky3970research, and I think that is a very, you know, positive add.3971    So I think the key is how we can make sure that, you know,3972one Federal dollar goes much, much further than that with a3973private investment on top of that.3974    Senator Lujan. Yesterday I reintroduced a piece of3975legislation called the ``Test AI Act'', which has bipartisan3976support, which would simply improve the Federal Government's3977capacity to test and evaluate in this area as well. So very3978much appreciate both your responses.3979    But this is just one of many steps I would argue that is3980needed to ensure that the United States stays ahead. Now,3981despite strong support across the country including from3982industry leaders here today President Trump is annihilating3983budgets for basic research, and there are questions abound by3984so many.3985    I will argue that this will destroy our Nation's3986competitive advantage. I simply just call on all my colleagues3987that we look at the investments to the National Science3988Foundation, National Institutes of Health, and Department of3989Energy, Office of Science.3990    Let us work together. If there is questions that we have3991let us find ways to address those. But let us ensure that these3992investments are making a positive difference so that we have3993more successes and more hearings celebrating what we are3994celebrating today.3995    Now, beyond your partnership with the Federal Government I3996would like to know more about how you partner with local3997communities when building out centers.3998    Data centers put a strain on energy and water resources.3999However, unlike other businesses they do not introduce many4000long-term jobs and economic benefits necessarily.4001    So, Mr. Smith, how many engineers do you have dedicated to4002model or hardware optimization to reduce energy use, and when4003you build a Center what initiatives do you have in place to4004reduce water use?4005    Mr. Smith. I do not know off the top of my head the number4006of engineers we have working on optimization but I would be4007happy to track down an answer and get it to you.4008    Water use is a huge priority especially, you know, in data4009centers, for example, in the southwestern United States and4010other countries around the world where water is in short4011supply.4012    If you look at our data centers today they run on liquid4013cooling. It is a closed loop system. The liquid is a4014combination of, frankly, water and other chemicals but4015basically once it starts running almost all of the water is4016recycled. So the amount of water that we consume is typically4017far, far smaller than what most people would estimate.4018    We also have a commitment to water replenishment. Our goal4019is to be water positive, meaning that we are providing more4020water to the community than we are consuming.4021    So, for example, across the United States today we have4022more than 90 water replenishment projects including one that4023focuses on the San Juan River in your state of New Mexico,4024which focuses on water security for the river.4025    So I think it is a good example of how we can play a4026responsible role in addressing an issue that is of growing4027importance.4028    Senator Lujan. I appreciate it.4029    Mr. Intrator, same question.4030    Mr. Intrator. Yes, I cannot answer the question of how many4031engineers we have focused on it but I will say that the ability4032to extract more computational power out of a given megawatt is4033of paramount importance to my company, to all of us in this4034room, and we spend an enormous amount of time integrating the4035most bleeding-edge technology, which is a step function more4036efficient in terms of its computational output than the legacy4037technology has historically done.4038    You know, so moving to liquid cooling has just been an4039incredible improvement in efficiency and, ultimately, we face4040this problem from, you know, within a given data center, within4041a given power envelope. How much can we move the computational4042resources forward, and that is really an important part of what4043we do.4044    Senator Lujan. I appreciate it.4045    Mr. Chairman, I have other questions I will submit into the4046record.4047    Mr. Moran did ask one question. Mr. Altman, you responded4048to it. But can you all just answer yes or no, is it important4049to ensure that in order for AI to reach its full prominence4050that people across the country should be able to connect to4051fast, affordable internet?4052    Dr. Su.4053    Ms. Su. Yes.4054    Mr. Intrator. Yes.4055    Mr. Smith. Yes.4056    Senator Lujan. Thank you. Appreciate it. I yield back.4057Thank you.4058    The Chairman [presiding]. Thank you.4059    Senator Lummis.40604061               STATEMENT OF HON. CYNTHIA LUMMIS,4062                   U.S. SENATOR FROM WYOMING40634064    Senator Lummis. Thank you, Mr. Chairman, and thank you all4065for coming today.4066    I really have been amazed at the outstanding progress that4067continues to be made in this field and I am already seeing4068people in Wyoming that are using ChatGPT or Claude to improve4069their businesses, whether it is health care or mining or oil4070and gas or education, ranching, even. I am just really excited4071about what this opportunity brings to America.4072    Now, as I see it, the world has presented us with two4073paths. On one hand, the EU has chosen to regulate first and ask4074questions later. The GDPR is already limiting European access4075to the most capable AI models.4076    On the other hand, China appears to be fast tracking AI4077development, standing up large amounts of energy very quickly4078in an attempt to outcompete America.4079    So, I would like to ask a few questions about how we can4080make sure we get the full benefit of this technology and4081accelerate its development.4082    So first question, over the past year we have seen many4083states including California and Texas consider their own AI4084frameworks, each one significantly burdensome in their own4085right. At the same time, our lead against China is shrinking to4086about only 6 months.4087    So, first of all, Mr. Altman, could you please sketch out4088what the world could look like if the U.S. were to have a4089patchwork regulatory framework and how that could impact our4090competitiveness?4091    Mr. Altman. I think it would be quite bad. I think it is4092very difficult to imagine us figuring out how to comply with 504093different sets of regulation and in many of these states there4094have been dozens of different bills proposed that I understand4095several of which could be passed. That will slow us down at the4096time where I do not think it is in anyone's interest for us to4097slow down.4098    One Federal framework that is light touch that we can4099understand and that lets us, you know, move with the speed that4100this moment calls for seems important and fine, but the sort of4101every state takes a different approach here I think would be4102quite burdensome and significantly impair our ability to do4103what we need to do and, hopefully, you all want us to do, too.4104    Senator Lummis. Does anyone disagree with Mr. Altman's4105assessment of a patchwork?4106    Thank you. I have some questions about the infrastructure4107that is going to be necessary to lead and compete in AI so my4108next questions are for our infrastructure providers, Mr. Smith4109and Mr.--is it Intrator?4110    Mr. Intrator. That is correct.4111    Senator Lummis. Intrator. Thank you. Could you elaborate on4112how current permitting processes have impacted your ability to4113rapidly deploy AI infrastructure? The more specific you can be4114the better.4115    Mr. Intrator. So a quick comment on the patchwork and then4116I will dive in here. The investment that we are making on the4117infrastructure side is enormous, and the idea that you can make4118an investment that could then become trapped in a jurisdiction4119that has a particular type of regulation that would not allow4120you to make full use of it is really very, very suboptimal and4121makes the decisionmaking around infrastructure challenging.4122    As far as the permitting goes, whenever this topic comes up4123the discussion around permitting is excruciating and it is4124excruciating from the ability to quickly build and to build4125large, and I think that is kind of from the data center forward4126without even beginning the discussion from the data center back4127through the energy infrastructure that is necessary to be able4128to power these large investments at the scale that make them of4129relevance to moving artificial intelligence forward. I am happy4130to spend more time kind of digging into more details but4131probably do that directly.4132    Senator Lummis. OK. And I will look forward to that4133conversation because I am worried about Wyoming's very clean4134natural gas being something your industry is concerned about4135because President Trump likes natural gas but President Biden4136did not.4137    And if you build huge data centers and another President4138comes along who is anti-natural gas that is a concern for you4139as you are deciding how to deploy capital.4140    Mr. Smith, do you agree?4141    Mr. Smith. Generally, I do. I mean, I would say we need4142consistency across administrations in this country. We need to4143find more opportunities for bipartisan agreement, and I will4144just highlight that in Cheyenne where we have long had a data4145center complex, you know, we do have backup generators that run4146on natural gas. So there are a variety of ways for us to put4147different energy supplies to good use.4148    Senator Lummis. Are you exploring small modular nuclear?4149    Mr. Smith. Yes, including with people in Wyoming.4150    Senator Lummis. Thank you.4151    Mr. Altman, I am pleased to hear you are releasing an4152open--oh, my time is up. Excuse me. It goes so fast.4153    Mr. Altman. I would love to talk to you about it another4154time. We are very excited about it, too.4155    Senator Lummis. Yes, thank you. I yield back.4156    The Chairman. Thank you.4157    Senator Rosen.41584159                STATEMENT OF HON. JACKY ROSEN,4160                    U.S. SENATOR FROM NEVADA41614162    Senator Rosen. Thank you, Chairman Cruz. So I am ready to4163push the button and, anyway, time does go by very fast. Thank4164you for having this hearing.4165    I really believe in the promise of AI. So exciting, and we4166have to ask the right questions in order to promote its growth4167on one hand, and how can explore and create these new4168possibilities and pathways and also how do we protect ourselves4169from bad actors or outcomes as best as we can know at the time.4170    And, Mr. Altman, thank you for spending some time with me4171yesterday. I look forward to continuing to work with you on4172this.4173    So I want to start a little bit today at DeepSeek, an4174adversarial AI, because in February I introduced bipartisan4175legislation with Senator Husted to prohibit using DeepSeek on4176government devices, and earlier this week Senator Cassidy and I4177introduced a bill that would expand those prohibitions to4178include Federal contractors.4179    So, Mr. Smith, what should our approach be to AI models4180that are developed in or by adversarial countries like the PRC?4181    Should we be concerned about our adversaries co-opting AI4182to promote a particular ideology, collect sensitive U.S. data,4183and how are you combating this threat?4184    Mr. Smith. Well, I think you can take the DeepSeek example4185and it illustrates it well, and I think it is just worth4186thinking about the fact that DeepSeek produced two things. They4187have a model that is an open source model and they have an4188application, the DeepSeek app.4189    At Microsoft we do not allow our employees to use the4190DeepSeek app. We did not put the DeepSeek app in our app store4191because of the kinds of concerns that you mentioned, namely,4192data going back to China and the app creating the kinds of4193content that I think people would say were associated with4194Chinese propaganda.4195    At the same time, because the model itself is an open4196source model it was possible for us to go in it, analyze it,4197and change the code in the model, which we and other people4198have the permission to do to remove the harmful side effects.4199    And so I think we have to always think about the different4200aspects of the technology. I will say put security first and4201then go forward from there.4202    Senator Rosen. Thank you. I think we all know that data is4203the real power in our current world. He or she or whomever owns4204the data really can control a lot of what we do.4205    But I want to move on and speak with you, Mr. Altman, about4206AI and anti-Semitism a little bit because earlier this year ADL4207released a report showing that several major generative AI4208models have perpetuated dangerous anti-Semitic stereo tropes4209and, sadly, conspiracy theories.4210    So, Mr. Altman, what steps is industry taking to ensure4211that AI models do not perpetuate anti-Semitism? Will you4212consider collaborating with civil society to create kind of a4213standard benchmark for AI related to anti-Semitism, use it as a4214form of evaluation, and then maybe we could use those for other4215forms of hate as well?4216    Mr. Altman. Of course, we do collaborate with civil society4217on this topic and we are excited to continue to do so.4218    We want our users to have freedom to use models in the way4219they want, but we also do not want them to be damaging to sort4220of the fabric of society or particular groups.4221    There will always be some debate and the question of free4222speech in the context of AI is novel and I think it is4223different than what we faced before.4224    We really do view these as tools for users one-on-one but,4225of course, we are not here to, you know, make horrible anti-4226Semitic products.4227    Senator Rosen. Thank you. I want to move on to--Senator4228Lujan talked about data center energy use, water use, something4229we are all really concerned about. I want to put on top of that4230a little bit about data center security, add that to the mix.4231    So last Congress I actually got a bill passed into law, my4232bipartisan ``Federal Data Center Enhancement Act''. It4233establishes cybersecurity and resiliency standards for Federal4234data centers.4235    And so to Mr. Smith, or--I am sorry, Dr. Su. Thank you.4236    Dr. Su, I want to ask you a little bit about hardware. Are4237there ways the hardware like the chips AMD designs, new chips4238that we are hoping to think about--I know my career as a4239software developer we just know things have gotten smarter,4240faster, and they just--the cooler they can be the better we can4241compute.4242    So how can we make our chips cooler? How can we make our4243data centers, our computing power, more secure? And I know4244interoperability is sometimes a factor. But can you talk about4245this a little bit?4246    Ms. Su. Sure. Thank you for the question, Senator.4247    Look, I think all of those things are extremely important,4248as you said. So in our part of the energy efficiency, you know,4249power constraints that we have from a chip standpoint, you4250know, our job is to continue to make our chips more and more4251efficient every year.4252    We have seen, you know, 30 times improvement over the last4253few years and we will continue to focus, you know, in that4254area.4255    And then to your comments about, you know, security and4256ensuring that our chips are secure and people are not somehow4257breaking into them, those are also very high priorities in our4258overall development cycle for future generation chips as well.4259    Senator Rosen. Thank you. I look forward to working with4260all of you again on these important issues.4261    Mr. Chairman.4262    The Chairman. Thank you.4263    Senator Sullivan.42644265                STATEMENT OF HON. DAN SULLIVAN,4266                    U.S. SENATOR FROM ALASKA42674268    Senator Sullivan. Thank you, Mr. Chairman. I want to thank4269the witnesses for the testimony today. I appreciate the4270Chairman calling this hearing, and I agree with Senator Cruz's4271opening statement about this is a matter of national economic4272and national security in terms of our race, however you want to4273call it--competition with China.4274    So I know this topic has been pressed but I want to just4275get--I want to dig down a little bit deeper. Do you agree with4276that, all of you?4277    I am just going to ask some quick questions, that this is a4278huge issue of national security, economic security, relative to4279China, and we as America need to win in that regard. Very4280important.4281    Everybody nodding their head. And then I know that it had4282been touched, but is the consensus among the witnesses that we4283are ahead right now but as a kind of tentative lead?4284    What would be--very quickly we will start with you, Mr.4285Altman. What is your assessment on that? I know you have4286already talked about it. I just want to set the context for4287some of the questions.4288    Mr. Altman. Yes, I believe we are leading the world right4289now. I believe we will continue to do so. We want to make AI in4290the United States and we want the whole world to get the4291benefit from that.4292    I think that is the strongest thing for the United States.4293I think it is also the right thing to do for all the people of4294the world.4295    And I really appreciate you all being with here--with us4296here today because I think we will need your help, and4297everything you are saying or almost everything you are saying4298sounds great.4299    Senator Sullivan. So as I ask this question I will ask if4300you guys think we are ahead, but then the key things when you4301say we need your help what would--very succinctly, sometimes we4302are not so smart up here--what would the key things be that you4303would need from the U.S. Government to help us maintain that4304lead and dominate this space, which is what I think we need to4305do?4306    Mr. Altman, again, to you real quick on that.4307    Mr. Altman. We have talked a little about infrastructure4308but I think we cannot overstate how important that is, and the4309ability to have that whole supply chain or as much of it as4310possible in the United States.4311    The previous technological revolutions have also been about4312infrastructure and the supply chain, but AI is different in4313terms of the magnitude of resources that we need.4314    So projects like Stargate that we are doing in the U.S.,4315things like bringing chip manufacturing, certainly, chip design4316to the U.S., permitting power quickly. Like, these are4317critical. If we do not get this right I do not think anything4318else we do can help.4319    On the model creation side, we have talked about the need4320for certainty in our ability to train and to have fair footing4321with the rest of the world to make sure we can remain4322competitive.4323    The ability to offer products under a reasonable, fair,4324light touch regulatory framework where we can go win in the4325market, because the products will be so key to the sort of4326feedback loops and making them better and better, and the4327ability to deploy them quickly and win at the product level in4328addition to the model and infrastructure and data area is4329really quite important.4330    The ability to bring the most talented people in the world4331here, the most talented researchers. We have a ton in the4332United States. There are more out in the world. We should try4333to get them all here, improving models here. I think those are4334some of the specifics.4335    Senator Sullivan. Good. That is very helpful.4336    Let me ask, Mr. Smith, two other ones that I want to touch4337on. I agree fully with Senator Lummis.4338    I am sure Senator Cruz has the same view. One of our4339comparative advantages over China, in my view, has to be4340energy--all of the above energy.4341    Hopefully, you have seen in Alaska we have a very large-4342scale LNG project that I think we are going to get off the4343ground here we have been working on for a long time.4344    We will have a hundred years supply of natural gas. So we4345want you guys all to come up to Alaska with your data centers.4346We have got cold weather. We got a lot of cold weather. We got4347gas. We got land. We got water. We got it all.4348    Mr. Altman. That is very compelling.4349    Senator Sullivan. So, yes, come on up. When this project is4350done, 100-year gas supply. A little colder than Texas.4351    So two questions that relate to our comparative advantage,4352Mr. Smith, and then any others who want to jump in.4353    Energy--do we think that is? I think it is. And then second4354it is, I think, somewhat of a disadvantage. It frustrates me.4355Maybe you guys do not see this.4356    We have had American finance companies, venture capital4357firms, banks, others, that, remarkably, all the opportunities4358we have in America are helping fund some of these projects in4359China.4360    I have been a real staunch opponent of Americans who have4361opportunities to invest in other places investing in Chinese4362AI, Chinese quantum, because we all know they are going to use4363that to help make their military more lethal. I mean, that is4364what they do.4365    I was reading recently about this Benchmark Capital. I do4366not know these guys but they evidently did a $75 million round4367for some--an AI company in China. Is that another problem as4368well, Mr. Smith?4369    Advantage energy problem--American companies financing our4370competition?4371    Mr. Smith. I would connect three things: energy, people and4372access to capital. The U.S. has huge resources in energy, but4373never underestimate the ability of China to build a lot of4374electrical power plants, maybe more and faster than any other4375country.4376    So we are better off going into that with the mindset that4377we have to keep up and not take anything for granted. But then4378I would say the number-one comparative advantage of the United4379States throughout the 50 years that have defined digital4380technology has been bringing the world's best people to our4381country and giving them access to venture capital, and we4382should continue to burnish both of those.4383    And I think you are right to ask where else is venture4384capital going. I will just say this. If we can keep bringing4385the best people to the United States and if we can keep4386educating the best people in the United States, I believe the4387money will be here to enable them to succeed.4388    But let us make sure we are continuing to bring the best4389people in the world and giving them the opportunity to build4390great companies here in the United States.4391    Senator Sullivan. And American venture capital funds4392funding Chinese AI, is that in our national interest?4393    Mr. Smith. I think there is a really good question about4394whether it is and I recognize that you all are quite rightly4395focused on that.4396    I will just keep saying, bring the people here. They will4397have access to the money and we will outcompete the world.4398    Senator Sullivan. Great. Thank you. Thank you, Mr.4399Chairman.4400    The Chairman. Thank you.4401    Senator Markey.44024403               STATEMENT OF HON. EDWARD MARKEY,4404                U.S. SENATOR FROM MASSACHUSETTS44054406    Senator Markey. Thank you, Mr. Chairman, very much.4407    I would like to talk about the environmental impact of4408artificial intelligence. Artificial intelligence can help us4409combat climate change by improving weather forecasts and4410enabling us to better predict power supply and demand. But4411designing and training and deploying AI models also poses real4412risks for our environment.4413    The massive data centers that are critical for AI4414development requires substantial amounts of electricity,4415putting stress on the grid and potentially raising costs for4416consumers.4417    These data centers also generate significant heat. Cooling4418them requires huge volumes of water, often in regions already4419facing droughts because of climate change, and some data4420centers have onsite backup diesel generators, which can cause4421respiratory and cardiovascular issues and can increase the risk4422of cancer for the surrounding community.4423    The truth is, we know too little about both the4424environmental costs and benefits of AI.4425    Mr. Smith, do you agree that it would be helpful for the4426government to conduct a comprehensive study on environmental4427impact of artificial intelligence?4428    Mr. Smith. Generally, yes. One study was just completed4429last December and I think it is worth updating periodically.4430    Senator Markey. Do you think it would be helpful for the4431government to convene stakeholders including from industry and4432academia to help better measure AI's environmental impact?4433    Mr. Smith. I think as well as many other things that need4434to be measured. Yes, I think there is a role to be played.4435    Senator Markey. Mr. Altman, do you agree that the Federal4436Government should help with studying and measuring the4437environmental impact of AI?4438    Mr. Altman. I think studying and measuring is usually a4439good thing. I do think that the conversation about the4440environmental impact on--of AI and the relative challenges and4441benefits has gotten somewhat out of whack.4442    I am hopeful that AI--you know, we have been trying to4443address climate environmental challenges unsuccessfully or not4444successfully enough for a long time. I think we need help.4445    I think AI can help us do that. We have proposed or we are4446in the process of building a 10-gigawatt facility and we have4447got another----4448    Senator Markey. My question is should the Federal4449Government be on an ongoing basis studying the impact of AI?4450    Mr. Altman. Sure, and I think you should use AI to help.4451    Senator Markey. So that is why this Congress introduced the4452``Artificial Intelligence Environmental Impact Act'' to study4453both the positive and negative consequences of AI.4454    As the technology continues to develop, as models become4455more efficient, and as we build out the infrastructure, we need4456to do it.4457    Yes, AI might find--may find a cure for cancer. It may, but4458AI also could help to contribute to a climate disaster. That is4459also equally true.4460    So we need to just keep both of those things right on the4461table, especially as the Trump administration is ignoring the4462fact that last year 94 percent of all new installed electrical4463generation capacity in the United States was wind, solar, and4464battery, and Trump has said he is going to destroy all4465incentives for continuation of that.4466    That is something you have to weigh in on to make sure he4467does not do that. So I look forward to working with you on4468that.4469    Now I want to turn to AI's impact on disadvantaged4470communities. After all, we are not just talking about using4471artificial intelligence to write e-mails or plan grocery lists.4472    We are talking about technology used to calculate a4473family's mortgage, screen an individual's job application, and4474determine a senior's medical care.4475    When used in these situations it is absolutely essential4476that AI-powered algorithms are free from bias and4477discrimination. So let us start with a simple question.4478    Mr. Smith, can algorithms be biased and cause4479discrimination?4480    Mr. Smith. They can, which is why we test to avoid that4481outcome.4482    Senator Markey. Same question, Mr. Altman. Can algorithms4483be biased and cause discrimination?4484    Mr. Altman. Of course.4485    Senator Markey. Of course. Mr. Altman, does OpenAI work to4486guard against such bias and discrimination in ChatGPT?4487    Mr. Altman. Of course.4488    Senator Markey. Of course. So I am glad to hear that4489because you recently stated that the government should not4490implement privacy regulations on AI but instead, quote,4491``respond very quickly as the problems emerge,'' and I am very4492deeply worried about that approach.4493    We do not need to wait and see if poorly tested and trained4494algorithms will harm marginalized communities. Artificial4495intelligence is already supercharging the bias and4496discrimination prevalent in our society. Biased and4497discriminatory algorithms mean black and brown families are4498less likely to obtain a mortgage.4499    It means people with disabilities are less likely to be4500recommended for a job opening and it means women are less4501likely to receive scholarships for higher education.4502    These are real harms that are happening right now. It is4503Congress' job to address these existing problems that come with4504the rapid development and deployment of AI and it is why I am4505the proud author of the ``AI Civil Rights Act'' which would4506ensure that companies review and eliminate bias and4507discrimination in their algorithms before developing and4508deploying them.4509    It has to happen simultaneously, and it will hold companies4510accountable when their algorithms cause harms against4511marginalized population.4512    I will be fighting to ensure AI does not stand for4513accelerating inequality in our Nation. All of the protections4514we have in the real world should be moved to the virtual world4515because the same discrimination--again, women, black, brown,4516communities with disabilities, LGBTQ community--are going to4517move online and we have to build in the protections against4518that bias right up front, because otherwise those same4519discriminatory practices will just migrate immediately and the4520responsibility of the industry will be to work with Congress to4521make sure we put those protections on the books.4522    Thank you, Mr. Chairman.4523    The Chairman. Thank you.4524    Senator Peters.45254526                STATEMENT OF HON. GARY PETERS,4527                   U.S. SENATOR FROM MICHIGAN45284529    Senator Peters. Thank you, Mr. Chairman, and thanks to all4530our witnesses. Thank you for being here.4531    It is an incredibly important topic and we appreciate your4532expertise.4533    As we are looking at making sure that the United States is4534the world leader in AI, certainly, we have been talking about4535supply chains and infrastructure and all of those aspects.4536    But one area that I want to particularly focus on is4537workforce and people to make sure that we have the talent4538there. That is why I authored the ``AI Scholarship for Service4539Act'' and the ``AI Training Act''.4540    Both of those were signed into law in 2022. Earlier this4541year I introduced my ``AI and Critical Technology Workforce4542Framework Act'' to continue the effort along those lines, and4543love to work with each of you as we look at other legislation4544necessary to make sure we have got the workforce trained to4545take advantage of this amazing technology.4546    I do want to do a shout out to the University of Michigan4547that actually became the first university in the world to4548provide generative AI tools for their entire student body to4549prepare them for the workforce of tomorrow. So I want to talk a4550little bit about the workforce.4551    Mr. Altman, when met last year in my office and had a great4552conversation you said that upwards of 70 percent of jobs could4553be eliminated by AI and you acknowledge the possible social4554disruption of this.4555    If that is happening we have to prepare for it. We are not4556going to stand in the way of the incredible opportunities here4557but if this is, indeed, going to occur, we have got to be4558thinking pretty deeply about how that will be managed and make4559sure that everybody can benefit from AI, not just a select few4560that benefit.4561    So talk to me about how you believe leaders in your4562industry can help mitigate job losses or deal with what could--4563as you described it last year, major social disruption?4564    Mr. Altman. The thing that I think is different this time4565than previous technological revolutions is the potential speed.4566Technological revolutions have impacted jobs and the economy4567for a long time.4568    Some jobs go away. Some new jobs get created. Many jobs4569just get more efficient and people are able to do more and earn4570more money and create more and that is great.4571    Over some period of time society can adapt to a huge amount4572of job change, and you can look at the last couple of centuries4573and see how much that has happened.4574    I do not know. I do not think anyone knows exactly how fast4575this is going to go, but it feels like it could be pretty fast.4576    The most important thing or one of the most important4577things, I think, we can do is to put tools in the hands of4578people early. We have a principle that we call ``iterative4579deployment''.4580    We want people to be getting used to this technology as it4581is developed. We have been doing this now for almost five years4582since our first product launch.4583    As society and this technology co-evolve putting great,4584capable tools in the hands of a lot of people and letting them4585figure out the new things that they are going to do and create4586for each other and come up with and provide sort of value back4587to the world on top of this new building block we have and the4588sort of scaffolding of society that is, I think, the best thing4589we can do as OpenAI and as our industry to be--sort of help4590smooth this transition.4591    Senator Peters. The idea we want to get to the point where4592AI is not displacing work but actually enhancing work, that4593people are more productive and doing things that we probably4594cannot even imagine what people will do. If we would look a4595hundred years ago we have jobs that no one----4596    Mr. Altman. You cannot imagine, and I do not think we can4597imagine the jobs on the other side of this. But even if you4598look today at what is happening with programming, which I will4599pick because it is sort of my background and near and dear to4600my heart, what it means to be a programmer and an effective4601programmer in May 2025 is very different than what it meant4602last time I was here in May 2023.4603    These tools have really changed what a programmer is4604capable of, the amount of code and software that the world is4605going to get.4606    And it is not like people do not hire software engineers4607anymore. They work in a different way and they are way more4608productive.4609    Senator Peters. Right. Right.4610    Dr. Su, we certainly talk a lot about open source AI but4611most of the conversation has been about software. However,4612making technology open and able to work together matters at4613every level, as you know, from chips that power the devices to4614the servers that are running behind the scenes.4615    So my question for you is, what are the benefits of open4616standards and system interoperability at the hardware level,4617not the software level, and what are the implications for4618innovation, national security, as well as resilience in the4619supply chain?4620    Ms. Su. Thank you for the question, Senator.4621    I think there are an incredible number of advantages to4622having an open ecosystem at the hardware and the software and4623the application level.4624    The idea is, you know, there is no one organization or one4625group that has all the good ideas and so enabling the ecosystem4626to work together so that you can choose the best solution at4627every level and then also optimization across a broad set of4628constituents is a good thing.4629    I think it is also very good from a security standpoint to4630ensure that, you know, again, there are many choices so that we4631are not dependent on a single ecosystem. So, you know, we4632continue to be very forward thinking in open standards as well4633as open ecosystems.4634    Senator Peters. So your model is an open model. I4635understand Nvidia is a closed model. Is there--what are the4636advantages and disadvantages? What should we be thinking about?4637    Ms. Su. I think the major advantage in an open model, and4638that is something that we very much support, is the idea that4639we can have innovation come from many different parties and,4640you know, whether that is hardware innovation so on the4641different chips or that is system innovation on putting all4642these things together.4643    And, you know, our goal is to make sure that we always have4644the best of the best and there are many different ways--many4645different parties that can contribute to that and that is why4646we are very forward leaning in terms of open ecosystems.4647    Senator Peters. Great. Thank you. Thank you, Mr. Chairman.4648    The Chairman. Thank you.4649    Senator Fetterman.46504651               STATEMENT OF HON. JOHN FETTERMAN,4652                 U.S. SENATOR FROM PENNSYLVANIA46534654    Senator Fetterman. Thank you, Mr. Chairman. Hello.4655    Mr. Smith, I am a big supporter of energy. For me energy4656security is national security and, of course, you know,4657renewables is about that. But, of course, other things as well,4658too--fossil. But also that also includes nuclear, of course.4659Nuclear is important.4660    And now then there is that kind of energy transition. My4661focus is also that I want to make sure that rate payers in4662Pennsylvania really are not hit too hard for throughout all of4663this.4664    Now, the Washington Post reported that increasing4665electricity demand for the data centers is going to raise up4666residential power bills, perhaps, as much by 20 percent.4667    Now, to me, that is really a concern for me and certainly4668for Pennsylvania families. Now, the data center, you know, has4669important jobs during construction and doing those things and4670that is a great thing, of course.4671    But they are not, I guess, long term. But the rate--those4672rates might last longer for that.4673    And now, I have been very--tracking the plan to reopen TMI.4674I mean, I had my own personal story is I had to grab my hamster4675and evacuate, you know, in that--during the meltdown in 1979.4676    You might consume--you might assume that I was anti-nuclear4677and that is not--it is a--I actually am very supportive of4678nuclear because that is an important part of the stack. If you4679really want to have--address climate change you cannot turn4680your back on nuclear, in my opinion.4681    But I know that is the power nuke--Microsoft's data center4682so now--and I really appreciate that. But if I am saying, now,4683if you are able to commit that the power purchase agreement,4684you know, it is not going to raise electricity for Pennsylvania4685families.4686    Mr. Smith. No, I think you raise a critical point. We have4687two principles that we follow when we are constructing these4688data centers.4689    Number one, we will invest to bring onto the grid an amount4690of electricity that equals the amount of electricity that we4691will use so that we are not tapping a constricted supply.4692    Number two, we will manage all of this in a way that4693ensures that our activity does not raise the price of4694electricity to the community.4695    And so I was describing earlier how if there is4696improvements that need to be made to the grid, as there often4697are, we will go to the utility commission. We will propose a4698change in the rate that we are charged so that we can pay for4699that improvement.4700    I just think it is a fact of life because I think you4701highlight something critical. There are a lot of jobs when the4702construction takes place. There are jobs afterwards but they4703are not as many.4704    One will wear out the welcome quickly if we tax, in effect,4705the neighborhood by asking everyone to pay more for their4706electricity because we have arrived. We get it. We know we have4707to be a good and responsible member of the neighborhood.4708    Senator Fetterman. Now, you know, one of the perks of being4709a senator is that--for me, anyway, I get an opportunity to meet4710people that have much more impressive kinds of jobs or careers4711that I have led.4712    And, now, Mr. Altman, now, this is going to--I am going to4713count this as a highlight.4714    Recently, like, I know the work that you have done you are4715really one of the people that are moving AI and now it is an4716opportunity. I was excited to meet you.4717    And now, people--you know, people ask me it is, like, if4718you are going to talk about AI and now I get to ask you, I4719mean, like, the literal--the expert.4720    You know, some people are worried about AI or whatever and4721I am, like, you know, what about the singularity so, you know,4722the people like that.4723    If you would address that, please.4724    Mr. Altman. Thank you, Senator, for the kind words and for4725normalizing hoodies in more spaces. I love to see that.4726    I am incredibly excited about the rate of progress but I4727also am cautious, and I would say, like, I do not know--I feel4728small next to it or something. I think this is beyond something4729that we all fully yet understand where it is going to go.4730    This is, I believe, among the biggest--maybe still trying4731to be the biggest technological revolutions humanity will have4732ever produced and I feel privileged to be here. I feel curious4733and interested in what is going to happen.4734    But I do think things are going to change quite4735substantially. I think humans have a wonderful ability to adapt4736and things that seem amazing will become the new normal very4737quickly. We will figure out how to use these tools to just do4738things we could never do before and I think it will be quite4739extraordinary.4740    But these are going to be tools that are capable of things4741that we cannot quite wrap our heads around, and some people4742call that--you know, as these tools start helping us to create4743next and future iterations some people call that singularity.4744    Some people call that the take off. Whatever it is, it4745feels like a sort of new era of human history and I think it is4746tremendously exciting that we get to live through that and we4747can make it a wonderful thing. But we have got to approach it4748with humility and some caution.4749    Mr. Fetterman. I mean, I just did--for me, it has been--I4750get a chance to ask questions to a lot of Edisons as well, too.4751    The kinds of things that you are all collectively involved4752are going to transform our society, and people will look back475350, 60 years ago and see what has happened. So to me, over to4754the Chairman. Thank you.4755    The Chairman. Thank you, Senator Fetterman.4756    Senator Klobuchar.4757    Senator Klobuchar. Thank you. Good thought, Senator4758Fetterman. Thank you.4759    So you guys have been sitting here so long that the Pope4760has been chosen.4761    [Laughter.]4762    Senator Klobuchar. We do not know who.4763    The Chairman. Congratulations, Amy.4764    Senator Klobuchar. The white smoke has come up.4765    The Chairman. Congratulations.4766    [Laughter.]4767    Senator Klobuchar. You are welcome. Probably would not4768work.4769    But in any case, it was--I left for some other things, came4770back because I had one more question that I wanted to ask and4771it is related to just the whole deep fake issue just because4772Senator Blackburn and Senator Coons and Senator Tillis and I4773worked on this really hard, and they are--Blackburn and Coons4774are in the lead of the bill.4775    But we have recently seen deep fake videos of Al Roker4776promoting a cure for high blood pressure, a deep fake of Brad4777Pitt asking for money from a hospital bed. Sony Music has4778worked with platforms to remove more than 75,000 songs with4779unauthorized deep fakes including voices of Harry Styles and4780Beyonce.4781    I recently--I mean, it is not just famous people. There is4782a Grammy-nominated artist from Minnesota. Talked to him about4783what is going on with digital replicas. So there is a real4784concern and it kind of gets at what Senator Schatz and I were4785talking about earlier with the news bill.4786    But I just wanted to make you all aware of this legislation4787because there were some differences on this and now we have4788gotten a coalition, including YouTube, supporting it as well as4789the Recording Industry Association, Motion Picture Association,4790SAG-AFTRA. So it is a big deal and I am hoping it is something4791that you will all look at.4792    But could you just comment? I would go to you, Mr. Smith,4793first about protecting people from having their likenesses4794replicated through AI without permission, and even if you all4795pledge to do it our obvious concern is that there will may be4796other companies that would not and that is why I think as we4797look at what these guardrails are the protection of digital--4798people's digital rights should be part of this.4799    Mr. Smith.4800    Mr. Smith. Yes. No, I think you are right to point to it.4801It has become a growing area of concern. You know, during the4802Presidential election last year both campaigns, both political4803parties, were concerned about the potential for deep fakes to4804be created.4805    We worked with both campaigns and both parties to address4806that. We see it being used in, really, ways that I would call4807abusive including of celebrities and the like.4808    I think it starts with an ability to identify when4809something has been created by AI and is not a genuine, say,4810photographic or video image, and we do find that AI is much4811more capable at doing that than, say, the human eye and human4812judgment.4813    I think it is right that there be certain guardrails and4814some of these we can apply voluntarily. We have been doing that4815across the industry.4816    OpenAI and Microsoft were both part of that last year, and4817there are certain uses that probably should be considered4818across the line and, therefore, should be unlawful, and I think4819that is where the kinds of initiatives that you are describing4820have a particularly important role to play.4821    Senator Klobuchar. And could you look at that legislation?4822    Mr. Smith. Absolutely.4823    Senator Klobuchar. Appreciate it. Mr. Altman, just the same4824question, same thing.4825    Mr. Altman. Of course, we would be happy to look at the4826legislation. I think this is a big issue and it is one coming4827quickly.4828    I do not believe--I think there are a few areas to attack4829it. You can talk about AI that generates content, platforms4830that distribute it, how takedowns work, how we educate society,4831and how we build in robustness to expect this is going to4832happen.4833    I do not believe it will be possible to stop the generation4834of the content. I think open source, open weight models are a4835great thing on the whole and something we need to pursue. But4836it does mean that there is going to be just a lot of these4837models floating around that can do this.4838    The mass distribution, I think it is possible to put some4839more guardrails in place and that seems important but I do not4840want to neglect the sort of societal education piece.4841    I think with every new technology there is some sort of--4842almost always some sort of new scams that come. The sooner we4843can get people to understand these, be on the lookout for them,4844talk about this as a thing that is coming and then a thing that4845is happening I think the better.4846    People are very quickly understanding that content can be4847AI generated and building new kinds of defenses in their own4848minds about it.4849    But still, you know, if you get a call and it sounds4850exactly like someone you know and they are panicked and they4851need help, or if you see a video that--like the videos you4852talked about, this, like, gets at us in a very deep4853psychological way and I think we need to build societal4854resilience because this is coming.4855    Senator Klobuchar. Mmm-hmm. It is coming, but we can4856there--there has got to be some ways to protect people's4857privacy rights----4858    Mr. Altman. We should do everything--for sure.4859    Senator Klobuchar.--and you have got to have some way to4860either enforce it--damages, whatever. There is just not going4861to be any consequences in that----4862    Mr. Altman. Absolutely. We should have all of that. Bad4863actors still do not always follow the laws and so I think we4864need an additional shield or whenever we can have them. But4865yes, we should absolutely protect that.4866    Senator Klobuchar. All right. Look forward to working with4867you on it. Thank you.4868    The Chairman. So I have to say Senator Klobuchar's question4869about fakes and AI fakes made me feel guilty because I did, in4870fact, tweet out an AI-generated picture of Senator Fetterman as4871the Pope of Greenland. So I am guilty of doing so, although it4872may not be a fake. It may be a real thing.4873    Senator Klobuchar. OK. Oh, whoa, parody is allowed under4874the law. Parody is allowed. That is different than what I am4875talking about but Senator Fetterman should respond.4876    Senator Fetterman. Or if it was AI.4877    Senator Klobuchar. I know.4878    The Chairman. It may be--it is a good shot, actually.4879    All right. I have a few more questions and then we will4880wrap up.4881    Mr. Altman, what has been the most surprising use for4882ChatGPT you have seen? What are applications that you are4883seeing that are surprising?4884    Mr. Altman. People message ChatGPT billions of times per4885day so they use it for all sorts of incredibly creative things.4886I will tell one personal story, which as mentioned earlier I4887recently had a newborn.4888    Clearly, people did it but I do not know how people figured4889out how to take care of newborns without ChatGPT. That has been4890a real lifesaver.4891    The Chairman. So I will tell you a story that I have told4892you before but my teenage daughter several months ago sent me4893this long, detailed text, and it was emotional and it was it4894was really well written and I actually commented. I am, like,4895wow, this is really well written.4896    She said, oh, I used ChatGPT to write it. Like, wait, you4897are texting your dad and you do not--it is something about the4898new generation that it is so seamlessly integrated into life4899that she is sending an e-mail, she is doing whatever, and she4900does not even--does not even hesitate to think about going to4901ChatGPT to capture her thoughts.4902    Mr. Altman. I have complicated feelings about that.4903    [Laughter.]4904    The Chairman. Well, use the app and then tell me what your4905feelings are.4906    Mr. Altman. OK.4907    The Chairman. Google just revealed that their search4908traffic on Safari declined for the first time ever.4909    Mr. Altman. It did not send me a Christmas card.4910    The Chairman. Will ChatGPT replace Google as the primary4911search engine, and if so, when?4912    Mr. Altman. Probably not. I mean, I think some use cases4913that people use search engines for today are definitely better4914done on a service like ChatGPT, but Google is like a ferocious4915competitor.4916    They have a very strong AI team, a lot of infrastructure, a4917very well-protected business, and they are making great4918progress putting AI into their search.4919    The Chairman. All right. So a question that I have spent a4920lot of time talking to business leaders, CEOs in the tech4921space, AI, and one question that I have asked that I get4922different answers on--and I am curious what the four of you4923say--how big a deal was DeepSeek?4924    Is it a major, seismic, shocking development from China? Is4925it not that big a deal? Is it somewhere in between and what is4926coming next?4927    And let us each of the four of you.4928    Mr. Altman. Not a huge deal. There are two things about4929DeepSeek. One is that they made a good open source model and4930the other is that they made a consumer app that, for the first4931time, briefly surpassed ChatGPT as the most downloaded AI tool,4932maybe the most downloaded app overall.4933    There are going to be a lot of good open source models and,4934clearly, there are incredibly talented people working at4935DeepSeek doing great research.4936    So I would expect more great models to come. Hopefully,4937also us and some of our colleagues will put out great models4938too.4939    On the consumer app, I think if a--if the DeepSeek consumer4940app looked like it was going to beat ChatGPT and our American4941colleagues' apps as sort of the default AI systems that people4942use that would be bad. But that does not currently look to us4943like what is happening.4944    Ms. Su. I would say it is somewhere in between, Chairman4945Cruz. When you think about what we have learned, what we4946learned is, you know, there are different ways of doing things.4947    So we have lots of incredibly innovative people in the4948United States. American models are, clearly, the best by far.4949However, when you have constraints that are placed there are4950other ways of doing things and I think we learned a few things4951in the process.4952    I think the open source nature of DeepSeek was one of the4953things that probably was most impactful in just terms of how4954much can be done in an open source type of model and open4955ecosystem.4956    But, clearly, the United States is leading and we need to4957continue, as we have said, to accelerate innovation and4958adoption as you started this hearing with.4959    Mr. Intrator. I think DeepSeek did a lot of things. One of4960the things that it did was it sort of raised the specter of4961China's AI capability to a much broader audience than was4962perhaps focused on it prior to that, right, and so you saw that4963kind of reverberate through the financial markets.4964    You saw, like, a broad-based reaction and suddenly everyone4965knows what DeepSeek is and the fact that China is not4966theoretically in the race for AI dominance but actually is very4967much a formidable competitor.4968    And so, you know, it was a starting gun in some ways for4969the broader population and kind of maybe the broader4970consciousness of the fact that that this is not a fait accompli4971and that we are going to have to work as America together to4972kind of propel our solutions forward. And so I think that was4973one of the lasting impacts that we will see from that.4974    Mr. Smith. I would say like Lisa that it was somewhere in4975between. It was not shocking. I mean, it was one of a number of4976startups that we were following in China that we saw as having4977the potential to be innovative in this space.4978    I do think there is a really interesting and important4979point that constraints encourage innovation in other ways and I4980just think one of the interesting facts about DeepSeek is that4981of their, say, 200 or more employees--that was the their size4982when they released these models--almost all of their employees4983by design were 4 years or less out of university.4984    They wanted to hire people that would not bring to their4985work traditional ways of doing things.4986    The Chairman. So the kids are taking over the world?4987    Mr. Smith. They do every generation.4988    [Laughter.]4989    The Chairman. Related to that--were you finished with that,4990Mr. Smith?4991    Related to that, we talked at the outset about the AI4992diffusion rule being rescinded, which I am glad. I think it was4993a bad rule. I think it was overly complex. I think it put on a4994number of our trading partners unfair restrictions and so I am4995glad the President is rescinding it.4996    That does not necessarily mean that there should be no4997restrictions and there are a variety of views on whether--what4998the rules should be concerning AI diffusion.4999    Nvidia has argued that we want American chips everywhere,5000even in China. Others have argued that we want to restrict at5001least the most advanced processors.5002    I am curious--each of the four of you what do you think the5003rule should be if anything is to replace the AI diffusion rule?5004    And, Mr. Altman, we will start with you.5005    Mr. Altman. I also was glad to see that rescinded. I agree5006there will need to be some constraints. But I think if our--if5007the sort of mental model is winning diffusion instead of5008stopping diffusion that directionally seems right.5009    That does not mean there is no guardrails. It does not mean5010we say, like, we are going to go build a bigger data center in5011some other country than the U.S. Our intention is to build our5012biggest and best data centers in the U.S. Do training in the5013U.S. Build models here. Have our core research here.5014    But then we do want to build inference centers with our5015partners around the world and we have been working with the5016U.S. Government on that. I think that will be good.5017    To this point that influence comes from people adopting5018U.S. products and services up and down the stack, maybe most5019obviously if they are using ChatGPT versus DeepSeek but also if5020they are using U.S. chips and U.S. data center technology and5021all of the amazing stuff Microsoft does, that is a win for us,5022and I think we should embrace that but make sure that, you5023know, the most critical stuff--the creation of these models5024that will be so impactful--that should still happen here.5025    The Chairman. Dr. Su.5026    Ms. Su. I think we would totally agree with the concept5027that some restrictions are necessary. This is a matter of5028national security as much as it is about AI diffusion.5029    That being the case, we were happy to see the rescinding as5030well and we view this as an opportunity to really simplify,5031right.5032    At the end of the day, you know, we have talked about the5033need to drive widespread adoption of our technology and our5034ecosystem. You know, simple rules that can be easily applied5035that really allow our allies to protect our technology while5036still utilizing the best that the United States has to offer I5037think is a good start in terms of where we are going and, you5038know, again, this is an area where I think the devil is in the5039details and it requires a lot of balance.5040    And so from an industry standpoint, you know, it is our job5041to put on the broader hat and work hand-in-hand with the5042administration and Congress to, you know, make our best5043recommendations so that it is a policy that has some stability5044as we go forward as well.5045    The Chairman. Mr. Intrator.5046    Mr. Intrator. So I will echo what Sam and Lisa said. But,5047you know, national security is paramount, and then once you5048have addressed the limitations around national security the5049opportunity to work with regulators to put together a5050regulatory framework beyond that makes a lot of sense, and the5051diffusion rule did not allow us that opportunity to participate5052fully enough to feel like we were going to come away with what5053would be an optimal outcome at this point.5054    The Chairman. Mr. Smith.5055    Mr. Smith. I think we have all discussed the right recipe.5056Simplify, eliminate these tier two quantitative restrictions5057that undermine confidence and access to American technology,5058but enable even the most advanced GPUs the country has to be5059exported to data centers that are run by a trusted provider,5060that meet certain security standards.5061    That means both physical and cybersecurity standards. That5062there is protection against diversion of the chips and there5063are precautions against certain uses, and that means two5064things.5065    One is that there are controls in place to ensure that,5066say, the PLA--the Chinese military--is not accessing and using5067these advanced models or advanced chips in a data center5068regardless of the country that it is in, and there are certain5069harmful uses that one should want to prohibit and preclude like5070using a model to create the next pandemic, a biological weapon,5071a nuclear weapon.5072    And I think that there is an approach that is coming5073together that can be retained and can move forward and that5074strikes the right balance.5075    The Chairman. OK. Final question for each of you. Would you5076support a 10-year learning period on states issuing5077comprehensive AI regulation or some form of Federal preemption5078to create an even playing field for AI developers and5079deployers?5080    Mr. Altman. I am not sure what a 10-year learning period5081means, but I think having one Federal approach focused on light5082touch on an even playing field sounds great to me.5083    Ms. Su. Aligned Federal approach with, you know, really5084thoughtful regulation would be very, very much appreciated.5085    Mr. Intrator. I agree with both my colleagues.5086    Mr. Smith. Yes, I think that builds, obviously, on the op-5087ed that you and Senator Graham published last year and I think5088giving the country time--your analogy, your example, was this5089worked for the internet.5090    There is a lot of details that need to be hammered out, but5091giving the Federal Government the ability to lead, especially5092in the areas around product safety and pre-release reviews and5093the like, would help this industry grow.5094    The Chairman. Well, I want to thank each of the witnesses.5095This was a very interesting hearing. It was informative. These5096issues matter.5097    You saw a great deal of interest on both sides of the aisle5098in this topic and so I appreciate--each of you are very busy5099and doing a lot of things and I appreciate your being here5100today.5101    Senators will have until the close of business on Thursday,5102May 15, to submit questions for the record and the witnesses5103will have until the end of the day on Thursday, May 29 to5104respond to those questions.5105    And with that, that concludes today's hearing. The5106Committee stands adjourned.5107    [Whereupon, at 1:13 p.m., the hearing was adjourned.]51085109                            A P P E N D I X51105111    Response to Written Question Submitted by Hon. Roger Wicker to5112                               Sam Altman5113Political and Ethical Decisions by AI Technology5114    Background: xAI was founded by Elon Musk on March 9, 2023, and5115develops technology similar to OpenAI, which Musk helped found5116alongside Sam Altman and others in 2015. Grok is xAI's flagship5117product, which runs on X (formerly Twitter), the social media platform.5118Musk, along with its developers, has expressed an intent to measure and5119potentially modify the political and ethical preferences embedded in AI5120systems. Studies have shown that popular AI models like OpenAI's5121ChatGPT tend to exhibit specific ideological leanings, particularly5122favoring environmental protection and expressing left-leaning,5123libertarian viewpoints.51245125    Question. Mr. Altman, President Trump has said that to maintain5126U.S. leadership in AI, we must develop systems that are free from5127ideological bias or engineered social agendas. You have said that you5128expect AI to be capable of superhuman persuasion well before it is5129superhuman at general intelligence. According to research recently5130published by xAI and Scale AI advisor Dan Hendrycks, AI systems exhibit5131significant left-wing biases in their value systems. What should be5132done to prevent superhuman persuasion by AI? Should superhuman5133persuasion by AI be banned? What are you doing to prevent superhuman5134persuasion in OpenAI's systems?5135    Answer. Our tools enable the freedom to learn, the freedom to5136create, and the freedom to innovate. We are accelerating knowledge,5137creativity, and free expression. Our systems have robust guardrails. We5138are transparent about how those guardrails work and we work hard to5139make sure we are enabling creativity and protecting everyone's freedom5140to use AI. Our models are specifically designed not to ``have an5141agenda'' which is outlined in our public documentation, such as the5142Model Spec, describing how our models work; the goal of an AI assistant5143is to assist humanity, not to shape it.5144                                 ______51455146  Response to Written Questions Submitted by Hon. Marsha Blackburn to5147                               Sam Altman5148    Question. The strength of American businesses has long depended on5149the enforcement of intellectual property (IP) rights. Recently, OpenAI5150has publicly argued that unless AI companies in the U.S. are permitted5151to broadly claim fair use of copyrighted content, the Nation will lose5152its competitive advantage to China in the AI sector.[1] Specifically, a5153recent comment letter by OpenAI stated ``there is little doubt that the5154PRC's AI developers will enjoy unfettered access to data--including5155copyrighted data--that will improve their models.'' This assertion5156stands in stark contrast to the long-standing American principles that5157prioritize IP protection as a driver of innovation and a safeguard5158against foreign competition. The United States' commitment to upholding5159property rights and the rule of law has been central to its leadership5160in technological development. Would you suggest that the U.S. adopt an5161approach to intellectual property rights more akin to that of China?5162    Answer. America's intellectual property laws have underpinned5163generations of American technology leadership, from the personal5164computer to the commercial internet. America's creative professionals5165have benefitted from the strong protections our laws give to creators,5166and America's technology innovators have benefitted from existing and5167longstanding IP doctrines such as fair use, which permits new5168technologies to interact with copyrighted works in transformative ways.5169This balanced framework has enabled the success of American AI, and the5170United States should stand by the existing American legal framework5171that has served our country so well.5172                                 ______51735174   Response to Written Questions Submitted by Hon. Maria Cantwell to5175                               Sam Altman5176AI Standards5177    The U.S. driving development of AI standards alongside the most5178advanced democracies in the world offers us an opportunity to set the5179``rules of the road'' for AI on the global stage.51805181    Question 1. In response to my question regarding NIST standards,5182you stated that NIST standards would not be necessary, but that they5183could be helpful in improving our competitiveness. Can you explain how5184you view NIST standards as helping the United States' competitiveness?5185    Answer. We know that in the race for 5G China took an active role5186in subsidizing and supporting companies in standard setting bodies.5187Their influence in voting bodies helped set the rules of the road for51885G. Similar dynamics are occurring with AI standards and NIST can play5189a role in supporting American companies to navigate these processes,5190especially for non-legacy technology companies.5191    In order for AI to benefit the world, technical standards would5192help countries build on U.S. technology and promote democratic AI. This5193is what our initiative, OpenAI for Countries, aims to achieve.51945195    Question 2. What standards would you like to see NIST develop and5196promote to improve U.S. competitiveness?5197    Answer. As models become more capable, it will be important to5198further the science of evaluations and metrics for measurement of such5199capabilities. NIST's measurement science initiatives (benchmarks, test5200beds, cryptographic validation, AI evaluations) can help further the5201development of these methods of evaluation and give industry a common5202vocabulary around safety and security.5203    Separately, various countries are proposing frameworks for risk5204management of frontier systems. NIST could play a role in harmonizing5205those across jurisdictions to help American products get to new5206markets.5207    Finally, agents will present a new challenge to how tasks and5208communication are conducted on the internet. NIST could help drive and5209establish consensus based industry technical standards to address this5210challenge.5211AI Safety5212    We are seeing a proliferation of deepfakes and other AI content5213that threatens the average person's ability to discern truth in media.5214And that's just one area in the field of AI that presents complicated5215safety questions. The U.S. AI Safety Institute plays a critical role in5216ensuring that AI systems are developed responsibly and that the most5217advanced models are fully tested. This is crucial for building trust5218and promoting wider adoption.52195220    Question 3. Do you support the work of the U.S. AI Safety5221Institute?5222    Answer. OpenAI has had a constructive partnership with the U.S. AI5223Safety Institute, focused on national security risks posed by dual-use5224AI capabilities. In our view, this is a good model for how a voluntary5225partnership between the Federal government and private sector can5226protect American national security and strengthen our economic5227competitiveness.5228Public Investment in Science5229    Government investment in fundamental science has been the backbone5230of American success in technology and innovation. If the United States5231wants to outcompete foreign adversaries, it cannot defund the National5232Science Foundation, National Institute of Standards and Technology5233(NIST), Department of Energy labs, or STEM education programs that5234power the AI workforce and ecosystem. Leadership in AI requires5235sustained public investment, not ill-conceived cuts that are not data5236driven.52375238    Question 4. How has your company benefited from or collaborated5239with the National Science Foundation, NIST or the Department of Energy5240Labs in artificial intelligence development?5241    Answer. OpenAI has greatly valued the opportunity to collaborate5242with U.S. public institutions, particularly the Department of Energy5243National Laboratories, including Los Alamos National Laboratory, in5244advancing the safe and beneficial use of artificial intelligence for5245scientific discovery.5246    Highlights of our collaboration include:52475248   Secure deployment of OpenAI models at Los Alamos National5249        Laboratory (LANL): In a first-of-its-kind partnership, we5250        enabled the deployment of our models on the Venado5251        supercomputing cluster, supporting high-assurance scientific5252        research within secure government environments.52535254   Wider engagement across the national lab ecosystem: We have5255        partnered with scientists from across the DOE laboratory5256        network, including hosting a ``1,000+ Scientists Jam Session''5257        where over 1,500 researchers at nine national labs explored how5258        AI can accelerate research. These collaborations provide mutual5259        learning: our models improve with real scientific feedback, and5260        lab scientists gain hands-on experience with frontier tools.52615262   Ongoing discussions on future-focused projects: We continue5263        to engage with the labs on mission-aligned areas such as energy5264        research, bioscience, and materials discovery, with the shared5265        goal of responsibly harnessing AI to support U.S. scientific5266        and technological leadership.52675268    Question 5. How will cuts to NSF funding impact your workforce and5269search for talent?5270    Answer. OpenAI benefits from and deeply values the strong5271scientific ecosystem fostered by U.S. institutions such as the National5272Science Foundation (NSF). NSF-funded programs play an important role in5273supporting the researchers, students, and discoveries that shape the5274future of artificial intelligence.5275    A healthy academic research environment:52765277   Expands the pool of AI-ready talent, including many of the5278        researchers we are proud to have hired.52795280   Strengthens foundational science, much of which underpins5281        progress in machine learning and adjacent fields.52825283   Supports broader innovation, ensuring that developments in5284        AI benefit from and contribute to the wider scientific5285        enterprise.52865287    We are committed to working with government and academic partners5288to ensure the U.S. remains a global leader in both talent and5289innovation.52905291    Question 6. What impact will cuts to Federal funding for science5292and research at universities have on U.S. competitiveness in AI?5293    Answer. The strength of America's research universities has long5294been a key driver of national competitiveness in advanced technologies,5295including AI. Public investment in university research plays a unique5296role in enabling both fundamental discovery and talent development.5297    In the context of AI:52985299   Many core innovations have emerged from university labs5300        supported by Federal grants.53015302   AI models themselves are increasingly being used to support5303        scientific discovery, amplifying the value of research5304        investment.53055306    Continued support for university-based research ensures that the5307U.S. remains at the forefront of scientific and technological progress.5308We are enthusiastic about the opportunity to continue contributing to5309this shared mission alongside academic and Federal partners.5310Energy Needs and R&D for Fusion Energy5311    The growing demand for electricity to power AI data centers is5312staggering. By some estimates, global electricity demand from data5313centers is projected to more than double by 2030 exceeding 9455314terawatt-hours (TWh). It will strain electric grids and energy5315providers.53165317    Question 7. What plan does your company have to meet energy needs5318for AI, and what investments are you making into non-fossil fuel5319sources of energy such as fusion?5320    Answer. We anticipate that AI's energy needs will incentivize5321substantial new investment in grid infrastructure and drive innovation5322in energy technologies. Advances in AI, including reasoning models,5323hold significant promise for scientific discovery, including in the5324field of abundant, affordable energy solutions. Indeed, our existing5325partnership to deploy reasoning models for use by the National Labs5326includes Lawrence Livermore, whose scientists were first in the world5327to demonstrate fusion ignition. Alongside others within the industry,5328we also will continue our work to find new ways to ensure our5329technology is as efficient as possible, including when it comes to5330energy consumption. Even as we continue to see promising research and5331innovation, we also remain focused on optimal use of available5332computing power, both in research and deployment.53335334    Question 8. With respect to fusion energy, how can the government5335partner with the private sector to scale fusion technology as it5336continues to develop?5337    Answer. The scaling laws are clear. American AI leadership is a5338function of energy, data, and chips. Government support for fusion5339research and pilots can be crucial, as companies look to identify5340viable paths to raising the capital they need for continued scientific5341progress and ultimately development at scale. As the government5342continues to support fusion research by private entities and at our5343national labs, we hope that our work with the National Labs as well as5344early explorations with a range of fission and fusion companies5345provides early indication of the role that AI can play in advancing5346energy abundance.5347                                 ______53485349    Response to Written Question Submitted by Hon. Amy Klobuchar to5350                               Sam Altman5351Topic: Workforce Development of Engineers5352    In your testimony you said that by the end of this year Open AI5353``will release AI powered tools that can handle sophisticated software5354engineering.'' I'm concerned we won't be able to grow senior engineers5355if AI replaces junior engineers.53565357    Question. How will you ensure Open AI grows the talent needed for5358future success?5359    Answer. We are very focused on fostering and training software5360engineering talent--including through the role our technology can play5361in bolstering American education at all levels and expanding computer5362science capabilities across our schools and all sectors of the economy.5363Our AI tools are highly complementary to existing software know-how and5364can dramatically increase the capabilities of small and large5365businesses. One example is how our tools can reduce a software5366engineer's manual work and allow those engineers to focus on more5367complex tasks, thereby strengthening their skills in important areas5368like critical thinking, creativity, and problem-solving.5369                                 ______53705371    Response to Written Questions Submitted by Hon. Brian Schatz to5372                               Sam Altman5373Future of Work5374    Question 1. How is your company taking advantage of the automation5375you're empowering to scale productivity without leaving workers behind?5376    Answer. Our AI tools are used in a wide range of ways at OpenAI,5377such as helping people with tasks like developing code, interacting5378with customers, and analyzing data. We see advanced AI as a way to help5379people and businesses gain new capabilities, increase innovation, and5380uplift productivity. As AI tools become increasingly capable, we think5381it's important to bolster education at all levels, from elementary5382schools through mid-career workers. Everyone should have the5383opportunity to benefit from AI.53845385    Question 2. Once OpenAI has ``generated orders of magnitude'' of5386returns on investments, do you believe the Federal government has any5387responsibility to make sure those benefits are distributed equitably5388across all Americans?5389    Answer. OpenAI is a relatively young company and we continue to5390invest at scale to develop new and more capable AI models, and we5391invest heavily in the infrastructure required to achieve our mission.5392We are not a profitable company at this stage, unlike more mature5393technology firms that have large, sustained profit margins. We believe5394the Federal government should ensure Americans have the freedom to5395access and benefit from AI as it advances.53965397    Question 3. OpenAI's capped profit structure was originally5398designed, in part, to mitigate the harms of workforce automation by5399using excess profits from AI to support those who lost their jobs.5400However, per your corporate restructuring announcement, you now intend5401to remove that cap. Do you still intend to support potential displaced5402workers under this new structure?5403    Answer. Earlier this month we reaffirmed our commitment to the5404OpenAI nonprofit having control over the organization. The previously5405existing for-profit subsidiary--originally structured as a ``capped-5406profit'' LLC--will be converted into a Public Benefit Corporation5407(PBC), and this new entity will remain under the control of the5408nonprofit. The PBC's mission will be the same as the nonprofit's, which5409is to ensure AGI benefits all of humanity. The new structure will allow5410us to strengthen our ability to attract capital, talent, and resources,5411while preserving our founding mission to make our services broadly5412available to all of humanity. We believe our mission of achieving safe5413and beneficial advanced AI will help people and businesses gain new5414capabilities, increase innovation, and uplift productivity.54155416    Question 4. Will OpenAI commit to developing clear standards not5417only for data quality, but for labor protections and responsible5418practices across the AI training data supply chain?5419    Answer. OpenAI has a strong track record of transparency around how5420our models are built and how we ensure they're safe and designed to5421prevent a wide range of potential harms. We will continue to work with5422the public and private sectors to provide insights and understanding as5423our technology advances. We also have made public our supplier code of5424conduct.5425Corporate Restructuring5426    In 2017, you said ``That's why we're a nonprofit: we don't ever5427want to be making decisions to benefit shareholders. The only people we5428want to be accountable to is humanity as a whole.'' In your previous5429testimony before the Senate, you explained the specific safeguards in5430OpenAI's structure that ensure it remains true to its charitable5431mission. On May 5, 2025, OpenAI announced that it would transition its5432for-profit operations to a Public Benefit Corporation (PBC), but that5433the nonprofit would retain control.54345435    Question 5. What mechanisms are in place to prevent mission drift5436and ensure that the PBC's actions align with OpenAI's foundational5437goals?5438    Answer. The nonprofit will control and be a large shareholder of5439the PBC. Both the nonprofit and the PBC will have the same mission--to5440ensure that AGI benefits all of humanity. We have described our plans,5441including our ongoing commitment to our mission, in this recent5442statement. Other AI labs, like Anthropic and xAI, are also PBCs, as are5443other purpose driven companies like Patagonia.54445445    Question 6. What criteria, metrics, or benchmarks will OpenAI use5446to evaluate whether its actions serve the public interest?5447    Answer. We will continue to be transparent about how our models5448work, including publishing a ``model spec'' that outlines how our5449models are designed and the safety guardrails incorporated into5450training. We also maintain a public preparedness framework, which5451details how we evaluate and mitigate potential AI harms.5452    In addition, our corporate structure is fundamentally designed to5453serve the public interest. As a Public Benefit Corporation (PBC), the5454PBC board of directors will hold a fiduciary duty to uphold the public5455benefit objectives outlined in our charter. These objectives are5456aligned with--and in fact identical to--those of the OpenAI nonprofit.5457This means that serving the public interest is not just a guiding5458principle, but a legal obligation embedded in our governance model.5459    By combining technical transparency, safety-focused metrics, and a5460mission-aligned corporate structure, we ensure that our actions remain5461squarely focused on advancing the public good.54625463    Question 7. Will OpenAI commit to regular public disclosures about5464its operations, decision-making processes, and AI developments?5465    Answer. We have a long track record of transparency and engagement5466on these issues, including published, in-depth research, system cards,5467safety specifications and testing information, ongoing research5468programs and academic partnerships, and disclosures about how our5469models and safety work are developed and implemented.54705471    Question 8. Do you still agree that the interests of your5472shareholders are not the same as the interests of the public, and might5473not always be aligned with America's security interests?5474    Answer. Our mission is to ensure that AGI benefits all of humanity,5475and that will not change. We believe that advancing democratic AI, led5476by the U.S. and like-minded nations and anchored by a commitment to5477freedom and democratic principles, is the best way to ensure both our5478mission and America's security interests. As we announced earlier in5479May, OpenAI was founded as a nonprofit, and is today overseen and5480controlled by that nonprofit. Going forward, it will continue to be5481controlled by that nonprofit. Our for-profit LLC, which has been under5482the nonprofit since 2019, will transition to a Public Benefit5483Corporation (PBC)--a purpose-driven company structure that has to5484consider the interests of both shareholders and the mission. The5485nonprofit will control and also be a large shareholder of the PBC,5486giving the nonprofit better resources to support many benefits. Our5487mission remains the same, and the PBC will have the same mission.54885489    Question 9. How will OpenAI manage potential tensions between5490profit-driven investor expectations and its nonprofit mission?5491    Answer. Our mission remains the same--to ensure that AGI benefits5492all of humanity. The nonprofit and PBC share the same mission.5493Safety and Security5494    Question 10. Do you agree that one of the board's core5495responsibilities is to ensure OpenAI's models are thoroughly tested5496before their release to ensure they won't harm the public?5497    Answer. We conduct extensive safety testing and outline both how5498our models are designed and the work we do to safeguard against5499potential risks. This process is outlined on our website and is5500discussed and detailed in extensive public documentation. Last year,5501the Board formed the Safety and Security Committee, an independent5502oversight committee focused on model safety and security. The Committee5503is briefed by company leadership on safety evaluations and exercises5504oversight over major model releases.5505    We also maintain a productive partnership with the U.S. AI Safety5506Institute that enables AI research, testing, and evaluation focused on5507national security risks.55085509    Question 11. Do you agree that employees should be encouraged to5510raise concerns about threats to the security of the United States?5511    Answer. OpenAI believes that open communication is essential to a5512successful work environment and that all employees should feel free to5513raise issues of concern without fear of reprisal. We have worked to5514foster a culture where people feel a responsibility to raise potential5515safety concerns and work to address them. This is encompassed in our5516Raising Concerns policy, which is made available to all OpenAI5517employees.55185519    Question 12. Does OpenAI maintain the commitments it made under the5520Biden Administration Voluntary AI Commitments?5521    Answer. OpenAI continues to prioritize the safe, secure, and5522transparent development and use of AI technology.55235524    Question 12a. If so, will OpenAI continue to maintain these5525commitments under its new PBC structure?5526    Answer. Regardless of our corporate structure, OpenAI will always5527be committed to the safe, secure, and transparent development and use5528of AI technology.55295530    Question 13. Does OpenAI maintain the commitments it made under the5531Frontier AI Safety Commitments?5532    Answer. OpenAI remains committed to fulfilling the voluntary5533Frontier AI Safety Commitments made at the Seoul AI Summit in May 2024.5534In advance of the Paris AI Action Summit in February 2025, we published5535an update showing our progress on the voluntary commitments made at5536Seoul and at previous AI summits, available here. [https://5537cdn.openai.com/global-affairs/paris-summit-update-on-voluntary-5538commitments-20250207.pdf]55395540    Question 13a. If so, will OpenAI continue to maintain these5541commitments under its PBC structure?5542    Answer. We will continue to prioritize the safe, secure, and5543transparent development and use of AI technology.5544Accelerating Scientific Research5545    Question 14. How is OpenAI expanding partnerships or tool5546development to accelerate breakthroughs in biotechnology and other5547scientific research, and what additional support do you need from5548Federal agencies to scale these efforts responsibly?5549    Answer. We are incredibly excited about how our AI models can5550accelerate scientific progress. We have established a strong5551partnership with the National Laboratories to further these efforts. We5552would continue to encourage public sector uptake of AI tools across5553institutions that work on science and health. And we strongly encourage5554efforts to incorporate AI into education at all levels, which will5555equip many more people to harness AI for scientific breakthroughs.55565557    Question 15. OpenAI models increasingly generate content used in5558scientific inference. What evaluation protocols or benchmarks do you5559believe are necessary to ensure that AI-generated results in scientific5560fields are robust, reproducible, up-to-date, and trustworthy?5561    Answer. Different fields have different approaches to this, but5562generally speaking, robust science should be reproducible and peer5563reviewed. This helps ensure the validity, significance, and originality5564of scientific work and helps improve the overall quality of research5565across various fields.5566United Arab Emirates (UAE)5567    Question 16. Please share why you believe the UAE is a trustworthy5568partner for increased collaboration on AI.5569    Answer. The UAE is aiming to be a world leader in integrating AI5570into their industry and society. They have spent years developing plans5571that bring together government, investor, private sector and5572educational institutions towards this end. We believe it's important5573that American companies don't sit on the sidelines as they proceed with5574their plans.5575    As we partner with countries under our OpenAI for Countries5576initiative we are ensuring that democratic values shape the future of5577AI. To do this securely, we need smart export controls that balance5578innovation and safety, while aligning nations around rights like free5579expression and safeguards against surveillance. We are working with the5580Federal government to ensure our international partnerships meet the5581highest standards of security and compliance.55825583    Question 17. What actions by the UAE would cause you to reconsider5584support for cooperation and collaboration on AI?5585    Answer. Our partnerships with foreign governments will only succeed5586by following the standards of security and compliance set forth by the5587United States government. Foreign governments will also be required to5588comply with our usage policies.5589Energy Consumption and Cost of Winning5590    Question 18. Do you support President Trump's efforts to expand5591coal power for AI data centers?5592    Answer. We support efforts to ensure AI infrastructure is powered5593by energy that is affordable, reliable, and scalable--goals that are5594critical to U.S. economic competitiveness and job creation. How best to5595achieve that, across a mix of energy sources, is a matter for5596policymakers and the broader energy sector.55975598    Question 19. How are you addressing the costs associated with new5599infrastructure development in the short-term, including in terms of5600water consumption, pollution, and climate impacts?5601    Answer. We take very seriously the impacts associated with AI5602infrastructure projects. We intend to actively engage with federal,5603state, and local officials, as well as communities, to responsibly5604manage water usage and climate impacts. Our aim is to ensure that our5605AI infrastructure projects not only support technological advancement5606but also contribute positively to sustainability and quality of life.56075608    Question 20. What plans do you have to source clean energy and to5609publicly report your companies' emissions?5610    Answer. We expect our near-term infrastructure projects to5611incorporate significant clean energy components, including nuclear5612energy, solar generation, wind and large-scale battery storage, as well5613as to comply with all applicable environmental and reporting5614requirements. We are also actively exploring how we can support5615accelerated timelines for advanced clean energy technologies, including5616advanced nuclear and fusion.5617                                 ______56185619   Response to Written Questions Submitted by Hon. Edward Markey to5620                               Sam Altman5621Comprehensive Impacts of Data Center Construction5622    Question 1. When planning for data center construction, does your5623company conduct a cradle-to-grave infrastructure study that includes5624wildlife, community, and pollution impacts during and beyond the5625operational lifespan of a data center? If yes, what have you learned5626from those studies? If no, why not?5627    Answer. In our ongoing site selection process, we are assessing5628sites in accordance with all applicable Federal and state environmental5629laws, as well as utilizing industry experiences, consultants, and best5630practices. We work closely with our infrastructure partners such as5631Oracle and Microsoft which have been responsible for these activities5632to date.5633Backup Energy Generation5634    Question 2. Does your company use backup diesel generators at any5635facilities?56365637    Question 2a. If yes, please provide a list of each facility where5638diesel generators are being used, along with the location, quantity,5639and type of generators.56405641    Question 2b. If yes, did your company consider the use of battery5642storage technology as an alternative to diesel generators? Please5643explain your decision process.5644    Answer. We are not currently utilizing backup diesel generators. In5645our site selection and design process, we will consider alternatives to5646diesel generators, including BESS technologies. We have battery backup5647power.5648Energy Mix5649    Question 3. Does your company utilize any on-site or colocated5650energy generation to power your data centers?56515652    Question 3a. If yes, please provide detail how much power comes5653from on-site and colocated energy generation.56545655    Question 3b. If yes, please list all on-site and colocated energy5656sources (e.g., renewable, nuclear, hydropower, gas-powered turbines,5657etc.) that are being utilized to power your data centers.5658    Answer. In developing our AI infrastructure, we will consider both5659on-site and utility power, including co-located energy sources.56605661    Question 4. How does your company ensure local ratepayers are not5662responsible for paying the cost of new energy infrastructure, such as5663transmission lines, needed to meet the data center's energy demand?5664    Answer. We intend to actively collaborate with utilities,5665regulators, and local governments to ensure that infrastructure costs5666associated with our data centers do not unfairly burden local5667ratepayers, maintaining fairness and affordability for all energy5668consumers5669Energy Consumption in AI Model Training5670    Question 5. How many GWh of energy do you estimate was used to5671train GPT-4.1 and any new AI models?5672    Answer. We are actively working to reduce energy consumption across5673all stages of AI development, from initial training to practical5674inference. We invest continuously in improvements to both hardware and5675software efficiency, collaborating closely with researchers,5676policymakers, and industry partners to establish stronger industry5677standards and credible, open benchmarks, such as those developed by ML5678Energy Initiative and Epoch AI.5679    As we explore new ways to enhance efficiency, we remain committed5680to thoughtful and responsible use of computing resources. Innovations5681like model distillation, which allows us to train large, comprehensive5682models that can subsequently be refined into smaller, highly efficient5683versions for targeted applications, are a key part of this strategy. As5684algorithms and AI technologies continue evolving, we anticipate ongoing5685improvements in both energy efficiency and resource conservation,5686including reductions in water usage.56875688    Question 6. Data centers require vast amounts of water for cooling.5689When water is at critically low levels, does your company continue to5690pull water for building cooling? Does it have a contingency for5691operating as to not put further stress on the water supply and5692potentially take limited resources from households, agriculture, or5693small businesses?5694    Answer. We are currently evaluating strategies to ensure5695sustainable water use, including in site selection, through advanced5696cooling technologies, and via operational adjustments and contingency5697plans that prioritize community water needs, ensuring responsible5698resource management and minimizing potential impacts on local5699agriculture, households, and small businesses.5700Government Partnerships5701    Question 7. Your company offers AI products specifically for the5702public sector, which are now used across Federal agencies and state and5703local governments. Given the especially heightened risks related to5704governments' use of AI--including the denial of rights or access to5705services and false or incorrect information about government benefits5706and programs--what additional steps have you taken to ensure that these5707tools are safe and effective to use in the context of government?5708    Answer. We provide advanced AI tools to consumers, businesses, and5709governments. Each of these use cases is different, but each requires5710safe and beneficial AI. We have outlined our safety and security work5711on our website and have published a ``model spec'' which provides5712extensive details about how we engineer safety in model development.57135714    Question 8. What protocols do you have in place to work with5715government agencies to rectify any harms or errors when they occur?5716    Answer. We solicit feedback from customers and partners, including5717government agencies. We don't decide how our models will be used by5718particular customers but work to ensure these models are highly capable5719and safe to deploy. We also provide a secure way to report harms5720through our Trust & Transparency reporting resources.5721Business Partnerships5722    Question 9. When you make your AI systems available to other users/5723deployers, what are the types of issues you agree on?5724    Answer. We have usage policies in place for our various products5725and tools, including our API, which businesses typically use as a base5726to build new products and services. Our usage policies are publicly5727available.57285729    Question 10. What information do you provide to those other5730parties/deployers?5731    Answer. We share publicly available information about how the5732models are trained, how they are designed to be safe, and our usage5733policies.57345735    Question 11. What step does your company take any steps to reduce5736the likelihood that, for instance, those downstream users don't use5737your services in ways that could harm people or that violate your terms5738of use?5739    Answer. We use a combination of automated systems, human review,5740and user reports to find and assess uses that potentially violate our5741policies. Violations can lead to actions against the content or a5742customer account, such as warnings, sharing restrictions, or5743ineligibility for inclusion in our services.5744AI Hallucinations5745    Recent reporting suggests that generative AI hallucinations are5746getting worse as the technologies become more powerful. Hallucinations5747can lead to great harm in certain scenarios, such as when assessing job5748applications, or even more dangerously in the national security5749context.57505751    Question 12. Do you agree that we need guardrails to ensure that AI5752tools are not used or misused in ways that could cause harm to people?5753    Answer. Yes and we have usage policies that are designed to protect5754against potential harms.57555756    Question 13. What steps is your company taking to address this5757issue?5758    Answer. We have usage policies that are designed to protect against5759potential harms.5760                                 ______57615762    Response to Written Question Submitted by Hon. Tammy Baldwin to5763                               Sam Altman5764    Question. You have spoken publicly on how you believe AI can be5765used to advance medical research and improve patient outcomes. However,5766I am concerned about sensitive personal information being compromised5767in the development of such models, and how systems may intrench bias.5768Can you share what role you envision AI having in health care and what5769protections need to be in place to safeguard patient privacy?5770    Answer. OpenAI has developed and implemented a range of privacy5771safeguards across the AI lifecycle, from model development to model5772deployment. We undertake extensive efforts at the training stages to5773limit the presence of personal information and decrease the likelihood5774that our models output sensitive personal information, and we provide5775users information and tools to address their privacy concerns. We are5776strongly committed to keeping secure any information we obtain from5777users or about users. These safeguards are outlined in our privacy5778policy.5779    We are already seeing our tools being used to accelerate scientific5780research and help fight disease. We have partnerships across medical5781research and clinical care. We have outlined our approach to working5782with healthcare providers, including privacy and security settings for5783patients. This is critical to ensuring AI tools are used safely in5784healthcare, where AI is a game changer.5785                                 ______57865787    Response to Written Questions Submitted by Hon. Jacky Rosen to5788                               Sam Altman5789Model Security5790    Question 1. It's essential we ensure the AI models we use do not5791become another cybersecurity vulnerability. Would voluntary5792cybersecurity standards for large AI models or high-risk models and the5793infrastructure they were trained on be helpful in establishing trust?5794    Answer. We are committed to continuously upleveling our5795cybersecurity practices, both from a security perspective (protecting5796our users, systems, and intellectual property) and from a safety lens5797(preparing for dual-use frontier capabilities, such as cyber, that have5798the potential to cause harm).5799    To ensure the security of our models and systems, we are leveraging5800our own AI technology to scale our cyber defenses and protect our5801users, systems, and intellectual property; partnering with third-party5802cybersecurity experts to rigorously test our cyber defenses through5803realistic red-teaming; and working to adopt industry-leading security5804practices such as zero-trust architectures and hardware-backed security5805solutions, together with our partners.5806    To ensure the safety of our models, we rigorously evaluate model5807capabilities consistent with our Preparedness Framework, and publish5808extensive documentation about how safeguards we have built into our5809models and approaches we use to ensure safety and security. We work5810closely with cybersecurity experts to conduct rigorous third-party5811assessments of our models, including with government agencies that are5812focused on combatting cyber risks. Lastly, we continuously monitor and5813disrupt attempts by malicious actors to exploit our technology.5814AI T&E Workforce5815    A key factor in ensuring the U.S. continues to lead the world in5816the AI race is by ensuring the AI we develop is the best and therefore5817the most trustworthy. Validating model outputs is an important step in5818establishing trust. Right now, however, the U.S. has neither the5819standards nor the trained workforce to evaluate AI models to establish5820that we can trust model outputs.58215822    Question 2. What should Congress consider to incentivize and grow5823the AI test and evaluation workforce?5824    Answer. We support congressional action to further educational5825opportunities to grow the AI workforce of the future. Additionally,5826OpenAI maintains a productive partnership with the U.S. AI Safety5827Institute that enables AI research, testing, and evaluation focused on5828national security risks. This partnership relies on skilled technical5829experts on the government side. More broadly, as AI tools are adopted5830across government, it will be important to promote AI skills and5831literacy across the Federal workforce, in order to ensure a capable5832workforce for AI development and implementation, and for the type of AI5833safety engineering and testing that you describe.58345835    Question 3. How can Congress support more interdisciplinary5836approaches to testing and evaluating AI? For example, how do we ensure5837a model being used in a healthcare setting has been evaluated both by5838experts in the model technology, but also experts in the healthcare5839setting in which it will be deployed?5840    Answer. Different AI use cases will require different approaches.5841We have worked extensively with regulated industries like financial5842services and healthcare to ensure our tools comply with their5843regulatory requirements. This is very important to ensuring AI is5844adopted across these sectors and that safety and security requirements5845are met.5846                                 ______58475848Response to Written Question Submitted by Hon. Lisa Blunt Rochester to5849                               Sam Altman5850    Mr. Altman, AI is becoming integrated into our critical economic5851and societal infrastructure, with McKinsey stating that long-term AI5852opportunity could be about $4.4 trillion in added productivity growth5853potential from corporate use cases.5854    But vendor lock in could be a real issue, where an AI vendor5855dramatically falls behind the competition and leaves its client with a5856vastly inferior product, which could threaten key industries the AI5857product operates in.58585859    Question. Do you have any plans or strategies regarding mitigating5860lock-in for your AI products operating in critical sectors, like the5861financial and medical sectors, to prevent potential lock-in effects5862that might harm these critical sectors and the folks therein?5863    Answer. AI is a highly competitive space, with lots of companies5864developing advanced model capabilities and new products. The state of5865the art is advancing rapidly. Firms typically work with multiple models5866and multiple developers, reducing the risk of lock-in to a particular5867provider. We think it's important that consumers and businesses are5868able to choose the best AI models and the best cloud infrastructure on5869which to run those models.5870                                 ______58715872     Response to Written Question Submitted by Hon. Todd Young to5873                              Dr. Lisa Su5874AI Public Awareness and Education5875    Winning the diffusion race not only requires providing a pathway5876for greater adoption of the technology and its applications into the5877general stream of commerce but also bolstering our public awareness and5878education of AI.58795880    Question. Dr. Su, would you like to comment on this, especially as5881it relates to building a workforce capable of solving more advanced5882scientific R&D challenges? If the workforce isn't here at home, where5883will it go and what are the consequences of that?5884    Answer. U.S. leadership in artificial intelligence is ultimately5885based on talent. The single greatest determinant of our long-term5886competitiveness is whether we can attract, train, and retain the people5887who will design, build, and govern these systems. Today, there are5888critical gaps across three main categories:58895890   Advanced AI research and systems engineering. There is a5891        significant shortfall of PhD-level talent capable of designing5892        and scaling frontier AI models. This includes expertise in5893        machine learning, algorithm optimization, and systems co-5894        design. Many of the leading researchers in this field are being5895        aggressively recruited--and retained--by overseas institutions,5896        often with the backing of state-directed strategies.58975898   Applied AI and deployment talent. We also lack enough5899        engineers who can safely and responsibly integrate AI into5900        critical sectors like healthcare, manufacturing, national5901        security, and logistics. These are cross-disciplinary roles5902        that require fluency in both AI and the domain in which it's5903        being applied. Bridging that gap is essential if we want AI5904        innovation to translate into broad-based economic benefit.59055906   Technical infrastructure and hardware specialization. AI5907        leadership requires leading in the underlying hardware stack--5908        everything from chip architecture and interconnect design to5909        advanced packaging and power management. The U.S. has a chronic5910        shortage of semiconductor engineers, firmware developers, and5911        skilled technicians. These roles are vital for making AI models5912        performant, efficient, and scalable.59135914    Addressing these gaps requires more than just graduate fellowships5915or university funding. A coordinated, national approach--from STEM5916education in K-12 to community college pathways, visa policy for high-5917skill immigrants, and public-private partnerships that give students5918and workers real exposure to cutting-edge AI development.5919    If the workforce is not in the United States, it will be in other5920countries, including strategic competitors, to their benefit and our5921detriment. This includes being able to attract and retain properly5922vetted foreign talent, many of whom seek to study, work and live here5923to contribute to our technology leadership.5924    If we want to lead the world in AI, we must lead the world in5925talent. That is the foundation--and right now, we are playing catch-up5926in too many areas.5927                                 ______59285929   Response to Written Questions Submitted by Hon. Maria Cantwell to5930                              Dr. Lisa Su5931AI Standards5932    The U.S. driving development of AI standards alongside the most5933advanced democracies in the world offers us an opportunity to set the5934``rules of the road'' for AI on the global stage.59355936    Question 1. How do NIST standards help the United States'5937competitiveness?5938    Answer. NIST's Risk Management Framework, including its comparisons5939with other AI frameworks and international standards, supports U.S.5940competitiveness because it was developed through a transparent public5941process which improves customer confidence. NIST has successfully used5942this framework in areas such as cybersecurity and AI, which AMD has5943relied upon.59445945    Question 2. What standards would you like to see NIST develop or5946promote to improve U.S. competitiveness?5947    Answer. While some standards, like chip design, are best driven by5948industry consortium, standards that protect national security, such as5949cybersecurity and silicon providence, are best when NIST partners with5950industry to promote U.S. competitiveness.59515952    Question 3. How has your company benefited from or collaborated5953with the National Science Foundation, NIST or the Department of Energy5954Labs in artificial intelligence development?5955    Answer. Through more than a decade of partnership with the5956Department of Energy, AMD now powers the world's two fastest5957supercomputers: Frontier, which went into operation at Oak Ridge5958National Labs in 2021, and El Capitan, which went into operation at5959Lawrence Livermore National Labs late last year. These systems are5960critical infrastructure for U.S. national security and scientific5961leadership, including the latest advances in drug discovery, medical5962research, climate research, hypersonic flight, and even training future5963generations of more capable Al models.59645965    Question 4. How will cuts to NSF funding impact your workforce and5966search for talent?5967    Answer. NSF funding for science and engineering research and5968education at U.S. colleges and universities provide formative training5969ground for the American semiconductor workforce pipeline. We support5970continued funding for these purposes to maintain U.S. AI leadership.59715972    Question 5. What impact will cuts to Federal funding for science5973and research at universities have on U.S. competitiveness in AI?5974    Answer. U.S. competitiveness in AI is based on five priorities: (1)5975accelerating U.S. chip and system innovations to keep our leadership in5976Al compute infrastructure where U.S. universities have played a key5977role; (2) open ecosystems that enable universities to directly5978contribute innovation; (3) U.S. research funding for advanced5979semiconductor manufacturing and packaging; (4) research investment do5980drive U.S. AI talent development; and (5) research in cyber techniques5981to strengthen U.S. defense capabilities. Federal funding for research5982related to these priorities at U.S. universities has been critical to5983maintaining our lead. To the extent less Federal funding is available5984for research related to these priorities at U.S. universities, the5985private sector will need to provide additional funding and support.59865987    Question 6. What are you most concerned with when it comes to your5988supply chains?5989    Answer. As a fabless company, AMD focuses on ensuring that the5990products it designs can be readily produced and delivered to customers.5991AMD consequently supports trade policies that allow those efforts to5992continue, including policies that do not impair the work underway to5993enhance semiconductor manufacturing capabilities in the United States.5994We must build a robust domestic supply chain for advanced semiconductor5995manufacturing and packaging. AI leadership depends on the ability to5996build complete, integrated systems. That means ensuring we have5997domestic capabilities in both wafer manufacturing at the most advanced5998nodes and next-generation packaging technologies, as well as the5999advanced system capabilities needed to bring it all together. This is6000an area where strong public-private partnerships are critical. The6001entire semiconductor industry is aligned on the need to work together6002and partner with the government to significantly scale U.S. chip6003production and advanced packaging capabilities here at home.60046005    Question 7. How will the higher costs from tariffs and potential6006supply chain disruptions impact your plans for building AI6007infrastructure?6008    Answer. AMD supports efforts to bring leading-edge manufacturing6009back to the U.S. We should be clear-eyed about the fact that building6010leading-edge semiconductor manufacturing will be a years-long project.6011AMD is encouraged by the new announcement by TSMC of an increase of6012$100 billion investment in the U.S. It is not viable to have 1006013percent of semiconductor manufacturing in one location. As an industry,6014we need to find ways to invest in the U.S. while building resiliency in6015our supply chain.60166017    Question 8. Do you anticipate any delays in construction or other6018work on AI infrastructure around the country? If so, where might these6019impacts hit the hardest?6020    Answer. This is a question best answered by parties building AI6021infrastructure in the U.S.6022                                 ______60236024   Response to Written Questions Submitted by Hon. Amy Klobuchar to6025                              Dr. Lisa Su6026Topic: Chips Supply Chain6027    The bipartisan CHIPS and Science Act was a landmark investment in6028domestic semiconductor R&D and manufacturing that I strongly support.6029Yet design and manufacturing are only part of the supply chain. For6030example, according to the Council on Foreign Relations, chips are6031typically sent to Southeast Asia for assembly, testing, and packaging.60326033    Question 1. What are additional key points in the chip supply chain6034that we need to focus on to bolster U.S. economic resiliency?6035    Answer. We appreciate the continued focus on strengthening6036America's semiconductor capabilities. Looking ahead, any future6037legislation offers an opportunity to address areas that are essential6038to our national and economic security but may have been underweighted6039in prior legislation. I'd offer three key priorities.60406041   First, U.S. leadership in chip design must be treated as a6042        national priority. Design is where technological6043        differentiation happens--where performance, efficiency, and6044        capability are defined. Other nations are investing6045        aggressively in this space, not just in manufacturing. A6046        forward-looking policy framework should include targeted tax6047        incentives and refundable R&D credits to ensure the U.S.6048        becomes a more attractive environment for advanced6049        semiconductor architecture, software tooling, and AI6050        accelerator innovation.60516052   Second, advanced packaging deserves far greater strategic6053        emphasis. The future of computing increasingly depends on6054        chiplet architectures and 3D integration. Yet today, as you6055        note, most of the high-end advanced packaging capacity resides6056        in Asia. This is a growing risk for both commercial and6057        national security applications. Focused investments in domestic6058        packaging R&D hubs and supply chain readiness can create6059        meaningful capability in an area where the U.S. still has time6060        to lead--but the window is closing.60616062   Third, talent must remain at the center of any long-term6063        strategy. The current shortage of semiconductor engineers,6064        software-hardware co-design specialists, and packaging6065        technicians is a structural challenge. We encourage continued6066        and expanded Federal support for workforce development6067        programs--from apprenticeship-style models and community6068        colleges to advanced university research partnerships that6069        align directly with industry needs.60706071    In short, we believe future action should be calibrated toward6072areas where strategic leverage is highest--design, packaging, and6073talent--and should be structured in a way that encourages broad6074private-sector investment without distorting the market.6075    We thank the Committee for its leadership and look forward to6076working together on policies that secure U.S. semiconductor leadership6077for the decades ahead.6078                                 ______60796080    Response to Written Questions Submitted by Hon. Brian Schatz to6081                              Dr. Lisa Su6082Securing American AI Systems6083    Question 1. Could you please detail what physical security and6084cybersecurity standards you have adopted, or will commit to adopting,6085to prevent IP theft and cyber disruptions by foreign adversaries?6086    Answer. U.S. companies have long-standing partnerships with6087manufacturers who treat chip technology IP with the utmost sensitivity6088to ensure customers continue using them for manufacturing.6089    For the physical security of AMD's products after manufacturing, we6090take our obligations under export control laws very seriously. As chip6091manufacturers, we implement strict Know Your Customer procedures and6092work closely with government partners to ensure compliance.6093    Effective enforcement ultimately depends on a strong partnership6094between industry and government, clear regulations, and rigorous6095oversight across the supply chain.60966097    Question 2. What investments, including know-your-customer6098processes, are you making in hardware security research and development6099to prevent your products from falling into the wrong hands?6100    Answer. AMD maintains a rigorous export control compliance program,6101which was developed in accordance with guidance from the U.S. Bureau of6102Industry and Security (``BIS''). Every single order must pass an export6103compliance review. We have well-established management policies on6104global trade compliance, which we regularly review and update, and6105distribute to all our employees. Each element of our program addresses6106specific requirements associated with U.S. and international6107regulations to make sure we export our products consistent with our6108obligations. Our compliance system also leverages software controls to6109block prohibited sales, such as to government sanctions lists, as well6110as controls reflecting bans against exports to certain locations and6111for specified end-uses.6112    Additionally, we leverage the latest technology and take a risk-6113based approach to our compliance efforts. Our goal is to know our6114customers, and we also gather information on indirect customers with6115point-of-sale records. Before we consider doing business with any6116party, we screen them for sanctions issues. We have deployed leading6117third-party resources--including the same ones trusted by the U.S.6118government--to scrutinize potential customers. These platforms6119continually study open-source data to see if it yields new6120intelligence, and we refresh our systems with their up-to-date6121findings. If red flags arise, our team conducts enhanced due diligence6122before taking any further steps.6123    At AMD, compliance is an enterprise-wide effort. All of our6124employees undertake mandatory compliance training on a yearly basis in6125order to stay current with their export control responsibilities. We6126also have a core GTC team of specialists around the world who are6127responsible for our export control policies, processes, and procedures.6128This team is located in six countries and collectively has decades of6129experience navigating export compliance regulations--and managing and6130executing export compliance policies--on a global scale.61316132    Question 3. Are these processes industry standards? How do we6133ensure that your competitors also exercise them?6134    Answer. AMD strives to have a best-in-class compliance program, as6135described above. Encouraging all companies to do the same and ensuring6136the government has sufficient resources to monitor compliance and apply6137the regulations in a consistent and transparent manner, will enhance6138the effectiveness of the current controls.6139PRC Deployment of AI6140    Question 4. What do you believe is the greatest national security6141threat posed by the People's Republic of China's deployment of AI6142systems?6143    Answer. AMD is a proud American company for over five decades. We6144want the U.S. to thrive, innovate and lead. That is good for us as a6145company and good for our employees, not just in the U.S. but around the6146world. As a company, AMD is not in a position to identify or prioritize6147national security threats.61486149    Question 5. What are your recommendations for addressing these6150threats?6151    Answer. The export control regulations describe the threats and6152impose restrictions, accordingly. As noted above, AMD's extensive6153compliance program is designed to comply with those restrictions,6154address the identified threats and prevent diversion, working closely6155with government partners to ensure compliance.6156Accelerating Scientific Research6157    Question 6. How can AMD's collaboration with U.S. national labs be6158deepened to ensure scientists across disciplines can access next-6159generation compute?6160    Answer. AMD's work in supercomputing illustrates the benefits of6161collaboration with the national labs. As mentioned above, through more6162than a decade of partnership with the Department of Energy, AMD now6163powers the world's two fastest supercomputers: Frontier, which went6164into operation at Oak Ridge National Labs in 2021, and El Capitan,6165which went into operation at Lawrence Livermore National Labs late last6166year. These systems are critical infrastructure for U.S. national6167security and scientific leadership, including the latest advances in6168drug discovery, medical research, climate research, hypersonic flight,6169and even training future generations of more capable Al models. In a6170similar vein, AMD stands ready to continue and enhance its6171collaboration with U.S. national labs in next-generation compute.6172                                 ______61736174    Response to Written Questions Submitted by Hon. Jacky Rosen to6175                              Dr. Lisa Su6176Competition Across the AI Ecosystem6177    It's essential that as we build a strong AI industry in the U.S.,6178we also focus on establishing a competitive one.61796180    Question 1. What should Congress consider to promote competition in6181the AI ecosystem and across the tech stack, such as interoperability6182requirements or securing access to computing power for researchers?6183    Answer. At AMD, we strongly believe that open ecosystems and6184interoperability are foundational to driving innovation, enhancing6185security, and fostering robust competition--especially in the fast-6186evolving AI sector.6187    Open ecosystems allow developers, researchers, and companies of all6188sizes to collaborate, build upon shared standards, and accelerate6189breakthroughs. By designing our platforms to support open-source6190software and industry-standard interfaces, we lower the barriers to6191entry for innovators, enabling a broader and more diverse set of6192participants to contribute. This democratization of access leads to6193faster iteration cycles and richer innovation pipeline because no one6194company, however large, has a monopoly on good ideas.6195    Interoperability ensures that technologies from different vendors6196can work together seamlessly. This not only gives customers more6197choice, but it also prevents vendor lock-in and encourages healthy6198market competition. From a national competitiveness standpoint, it6199strengthens the entire AI supply chain by creating resilience,6200redundancy, and flexibility.6201    Critically, openness also promotes transparency and security. Open6202standards and community-driven software allow vulnerabilities to be6203identified and resolved more quickly. This collaborative scrutiny6204results in more secure systems than those built in closed silos.6205    In short, interoperability and open ecosystems are not just good6206engineering practices--they are strategic imperatives. They catalyze6207innovation, enhance trust, and ensure that the U.S. remains at the6208forefront of global technology leadership in AI.6209    As we consider how to foster a more inclusive and competitive AI6210ecosystem, it's essential to emphasize that this is not just an6211innovation issue--it's also a national security imperative.6212    Open ecosystems and interoperability are particularly useful to6213help ensure that we are not concentrating risk, capability, or6214decision-making power in the hands of a few companies or platforms. In6215a world where AI systems will increasingly be used to support defense,6216critical infrastructure, and intelligence operations, resilience and6217diversity in the ecosystem are just as important as raw capability.6218AI Skills6219    Question 2. Both of you mentioned the importance of digital skills6220in your testimony. Can you discuss how it might hurt the U.S.'s ability6221to compete with China if we don't leverage congressionally-mandated6222Federal programs like those created under the Digital Equity Act, which6223were explicitly designed to help Americans build digital skills, like6224teaching seniors, small businesses, and veterans how to use AI?6225    Answer. AMD recognizes the importance of digital skills for the6226U.S. to compete and lead on the international stage. Talent must remain6227at the center of any long-term strategy to succeed. The current6228shortage of semiconductor engineers, software-hardware co-design6229specialists, and packaging technicians is a structural challenge. We6230encourage continued and expanded Federal support for workforce6231development programs--from apprenticeship-style models and community6232colleges to advanced university research partnerships that align6233directly with industry needs.6234    We must invest in talent and ensure our national strategy for STEM6235education and workforce training. The private sector can certainly do6236more, including expanding university partnerships, investing in6237reskilling programs, and developing the cross-disciplinary talent6238required for success. We should incentivize companies to increase their6239most critical Al R&D efforts. We should make America the absolute best6240place for Al talent in the world.6241    To that end, we recognize education holds immense power in shaping6242the leaders and innovators of tomorrow, and we are passionate about6243enhancing digital literacy and preparing students for the demands of6244the 21st century workforce. This is why we partner with schools,6245educators, and local nonprofit organizations to outfit AMD Learning6246Labs with AMD processor-based equipment, helping empower teachers and6247inspire students to pursue STEM education.6248    As additional examples of AMD's commitment to skills development6249and training the leaders of tomorrow, AMD supports the Ann Richards6250School for Young Women Leaders, a unique public school in Austin,6251Texas, which provides out-of-the-box education strategies and6252enrichment opportunities that incorporate real-world, hands-on projects6253that prepare and equip students to tackle big problems with big ideas.6254High school students complete a college-to-career pathway in STEM6255fields where women are historically underrepresented. In the university6256context, AMD University Program offers professors and lecturers free6257software licenses, hardware donations, and educational resources to6258support classroom teaching in digital design, embedded systems,6259computer science and AI.6260    Beyond the classroom, AMD also welcomed our first cohort of6261veterans in late 2023 through the Hiring Our Heroes (HOH) Corporate6262Fellowship Program, strengthening our workforce. The HOH Program,6263developed by the U.S. Chamber of Commerce, immerses transitioning6264service members, veterans, military spouses, and military caregivers in6265the civilian workforce, creating economic opportunity by building their6266experience and increasing the possibility of a job in the industry.6267    Through private sector initiatives, such as those at AMD, and the6268support of Federal programs, we can pave the way for the digital skills6269literacy for the U.S. to compete and succeed on the international6270stage.6271                                 ______62726273   Response to Written Questions Submitted by Hon. John Fetterman to6274                              Dr. Lisa Su6275    Question 1. Dr. Su, data centers use far too much energy--U.S.6276government data from the Lawrence Berkley National Lab expects that6277they'll use up to 12 percent of all U.S. electricity by 2028.\1\ While6278my Republican colleagues have been busy overturning important energy6279efficiency standards, it's more critical than ever for American6280industry to lead on energy efficiency and renewable energy. What role6281can hardware manufacturers play in improving data center energy6282efficiency?6283---------------------------------------------------------------------------6284    \1\ https://newscenter.lbl.gov/2025/01/15/berkeley-lab-report-6285evaluates-increase-in-electricity-demand-from-data-centers/6286---------------------------------------------------------------------------6287    Answer. We agree that, as AI systems grow more capable, the demand6288for compute--and the energy to power them--is rising quickly. Without6289thoughtful planning, we risk that pressure falling unfairly on everyday6290Americans who share the grid.6291    One of the most important ways that we can address this energy6292concern is at the chip level. Improving compute per watt--that is, how6293much useful work we get out of each unit of energy--is essential. As a6294chip design company, we see this as a central part of our mission.6295Every generation of AI hardware must be significantly more energy6296efficient than the last. That's not just good engineering--it's good6297economics and good energy policy.6298    We also need to be smarter about where and how AI workloads are6299run. More efficient chips, paired with intelligent system design, can6300dramatically reduce the power needed to train and deploy these models--6301especially at scale.6302    At the same time, industry needs to partner with utilities and6303policymakers to ensure the costs of growth are not passed on to6304households. That means investing in clean energy, modernizing grid6305infrastructure, and being part of the solution--financially and6306operationally.6307    AI leadership shouldn't come at the cost of affordability or6308sustainability. If we lead on efficiency--from the chip level all the6309way up--we can deliver more innovation with less energy, and make sure6310the benefits of AI are shared broadly.63116312    Question 2. I appreciate that you referenced technological and6313process efficiency in your testimony. Specifically, what is AMD doing6314to improve the energy efficiency of its chips for data center use?6315    Answer. AMD is highly focused on improving the energy efficiency of6316its chips for data center use. AMD takes a holistic approach to energy6317efficient design, balancing advancements across the many complex6318architectural levers that make up chip design, incorporating tight6319integration of compute and memory with chiplet architectures, advanced6320packaging, software partitions, and new interconnects. One of our6321primary goals across all of our products is to extract as much6322performance as possible while balancing energy use.6323    In addition, AMD is not only working on improving the efficiency of6324its own solutions, but working with partners and the larger ecosystem6325to optimize virtually every aspect of the AI pipeline. Optimization of6326its processing units and myriad of connectivity technologies, which6327link chips, systems and racks, will all help enhance efficiency, along6328with quantizing models, improving software, and tweaking algorithms.6329AMD's holistic approach to optimizing power efficiency means6330continually addressing every link in the virtual AI chain to maximize6331performance-per-watt.6332    This is an important consideration because it means the power and6333energy requirements of products when they initially hit the market,6334typically improve over the lifetime of the product. AMD has made6335significant efficiency gains year over year, and supercomputers built6336using AMD technologies have earned top rankings on the GREEN500. At one6337point, the AMD-powered Frontier TDS (test and development system) at6338Oak Ridge National Labs actually topped the GREEN500 list. The GREEN5006339ranks supercomputers from the TOP500 list, in terms of energy6340efficiency.6341    One of the key areas where significant efficiency gains are6342possible relate to data movement. The largest AI models require huge6343amounts of data. As bits move from the tiny register files inside GPUs6344or accelerator chips, to cache memory, out to High Bandwidth Memory,6345and to the CPU, and so on, energy consumption grows exponentially.6346Keeping as much data as close to the accelerator as possible is6347paramount to maximizing energy efficiency. It's why, for example, AMD6348continues to increase the amount of cache and memory on its Instinct6349accelerators generation-on-generation, and why the company continually6350explores ways to optimize how the data is actually processed.6351    If we look at typical, large-scale AI system today, roughly half6352the total power required to run the system is consumed by the GPU's6353high bandwidth memory (HBM), but the other half is comprised of CPU,6354scale-up and scale-out networking, and various things like cooling and6355other data center facility overhead. AMD's goal is to maximize system-6356level performance, while also minimizing total power consumption, not6357just from its chips, but from everything around them in the data center6358as well.6359    While the significant amounts of compute resources required for AI6360today are a concern, AMD is working hard to maximize the efficiency of6361its platforms and meet its related efficiency goals.6362                                 ______63636364   Response to Written Questions Submitted by Hon. Maria Cantwell to6365                            Michael Intrator6366AI Standards6367    The U.S. driving development of AI standards alongside the most6368advanced democracies in the world offers us an opportunity to set the6369``rules of the road'' for AI on the global stage.63706371    Question 1. How do NIST standards help the United States'6372competitiveness?6373    Answer. NIST special publications, developed through broad6374stakeholder consultation, offer American companies specific and6375adaptable guidance for creating frameworks to develop and deploy AI6376technologies. For example, the NIST Cybersecurity Framework assists6377U.S. firms by providing a common vocabulary and voluntary guidelines6378for information security and cyber risk management.63796380    Question 2. What standards would you like to see NIST develop and6381promote to improve U.S. competitiveness?6382    Answer. By providing the technical foundation and establishing6383common vocabulary for industry-led standards development, NIST can help6384ensure that U.S. security approaches shape global benchmarks.6385Public Investment in Science6386    Government investment in basic research has contributed6387significantly to U.S. advances in scientific discovery, innovation and6388U.S. competitiveness. Continued investment in the U.S. research complex6389is essential to maintaining U.S. economic strength and national6390security. This is also the case with AI given the need to continue to6391develop our high performance computing capabilities and next generation6392energy technologies required to power the data centers of the future.63936394    Question 3. How has your company benefited from or collaborated6395with the National Science Foundation, NIST or the Department of Energy6396Labs in artificial intelligence development?6397    Answer. As of May 2025, to the best of its knowledge CoreWeave has6398not formally collaborated with the Department of Energy Labs, the6399National Science Foundation, or NIST. However, we do look forward to6400opportunities to collaborate with the Department of Energy on its6401efforts regarding potential data center development on Federal lands.64026403    Question 4. How will cuts to NSF funding impact your workforce and6404search for talent?6405    Answer. CoreWeave supports NSF's important role in workforce6406development by training U.S. workers in emerging technologies. For6407example, NSF's Experiential Learning in Emerging and Novel Technologies6408program aims to provide experiential learning opportunities for6409individuals interested in career pathways in key technologies such as6410artificial intelligence, semiconductors, advanced manufacturing, and6411more. These types of programs bolster U.S. competitiveness and ensure6412that the U.S. has a pipeline of talent to support AI infrastructure and6413AI development.6414    A skilled and trained workforce is vital for the stability and6415expansion of AI data centers--which rely on specialized data center6416technicians, network and electrical engineers, cybersecurity6417professionals, and project managers. CoreWeave supports efforts to6418train the domestic workforce comprised of the skilled workers required6419to meet the growing AI demand and to accelerate AI innovation.64206421    Question 5. What impact will cuts to Federal funding for science6422and research at universities have on U.S. competitiveness in AI?6423    Answer. Science and research universities play an important role in6424maintaining global leadership in AI. Researchers and students6425collaborate with the technology industry on cutting-edge projects to6426shape the future landscape of AI innovation. A unique American6427advantage in the AI race is our ability to support our researchers and6428students at universities, as well as to create effective partnerships6429between academia, industry, and government.6430    CoreWeave is deeply committed to supporting AI research, science,6431and education, and public-private collaborative partnerships.6432    For example, CoreWeave is proud to be a founding partner of the New6433Jersey AI Hub, along with Microsoft, Princeton University, and the New6434Jersey Economic Development Authority. The founding partners will6435collectively invest over $72 million to support the long-term success6436of the NJ AI Hub, which focuses on research and development efforts,6437applications of AI in several industry sectors, and AI workforce6438development and education.6439Tariff Impacts6440    High tariffs and overreaching export controls, especially that are6441not well coordinated with the private sector and U.S. allies, have the6442potential to disrupt supply chains and raise costs for U.S. companies.6443That makes building AI infrastructure like data centers, chips, power6444plants, and grid modernization more expensive.64456446    Question 6. What are you most concerned with when it comes to your6447supply chains?6448    Answer. CoreWeave invests billions of dollars in the equipment6449which powers AI. Like any business which purchases long-lived capital6450assets, CoreWeave relies on cost and policy certainty to plan its6451business and make critical investments in the U.S. This enables us to6452provide predictable pricing to our customers.6453    One of our top concerns is ensuring strategic investment stability6454and predictable prices in the supply chain. We believe it's important6455to continue to maintain robust supply chain relationships with like-6456minded and reliable international strategic partners, including6457chipmakers, original equipment manufacturers, and software providers,6458to continue to scale AI data center operations in the U.S. and lead in6459the global AI race.64606461    Question 7. How will the higher costs from tariffs and potential6462supply chain disruptions impact your plans for building AI6463infrastructure?6464    Answer. Volatility in the global supply chain for critical6465components, such as advanced semiconductors and networking equipment,6466can raise costs or disrupt deployment timelines, adversely affecting6467American companies' ability to rapidly scale the AI capabilities6468necessary to meet the requirements of leading enterprise companies and6469AI labs.6470    Acquiring the necessary high-performance components to power AI6471workloads requires managing a complex global supply chain and6472maintaining robust supply chain relationships. Continued engagement6473with leading global suppliers and strategic partners is vital to6474ensuring the continued operation, expansion, and rapid deployment of6475U.S. AI infrastructure and to uphold U.S. competitiveness.6476    We are focused on a supply chain strategy that maintains robust,6477resilient access to the critical components we need to continue to6478develop AI infrastructure. We hope the ongoing discussions with trading6479partners will provide the clarity and certainty American companies6480require to make the large scale capital necessary to build the AI6481infrastructure at the scale and urgency this moment requires.64826483    Question 8. Do you anticipate any delays in construction or other6484work on AI infrastructure around the country? If so, where might these6485impacts hit the hardest?6486    Answer. No, we do not anticipate delays in construction or machine6487(e.g., GPU, storage, networking) work on AI infrastructure. CoreWeave6488does encounter challenges that we factor into our planning and6489construction processes. Acquiring utility power can involve long lead6490times for switch gear and large transformers due to high demand and the6491availability of critical components. Access to both data center6492equipment and underlying materials, such as steel or concrete, requires6493close collaboration with data center operators and supply chain6494vendors.6495    Labor constraints can pose another challenge, given the high demand6496for data center infrastructure and the limited pool of skilled6497tradesmen and technicians. Even in markets experienced with data center6498zoning and permitting, government administrative capacity to manage6499these processes can be a limiting factor.6500                                 ______65016502    Response to Written Question Submitted by Hon. Amy Klobuchar to6503                            Michael Intrator6504Topic: Competition in Cloud Infrastructure6505    The Federal Trade Commission has raised concerns that startups face6506significant barriers to entry into artificial intelligence (AI) markets6507because large technology firms have more access to troves of data,6508individuals with the necessary expertise, and computational resources.65096510    Question. Can you describe the importance of ensuring that6511innovative AI startups are able to thrive in markets alongside large,6512established tech companies? As a relatively new entrant into the AI6513infrastructure market, what factors are important to ensure a6514competitive ecosystem?6515    Answer. CoreWeave's infrastructure is the only cloud platform that6516was purpose-built for AI and Machine Learning (ML) workloads at the6517maximum performance and efficiency. CoreWeave's success to date6518demonstrates that cutting-edge innovation, rapid execution, and6519unparalleled performance has enabled the company to compete with6520incumbent platforms.6521    Our recent growth shows that the market can reward companies that6522differentiate based on performance, delivering specialized solutions to6523customers' specific needs. Through our proprietary software6524capabilities, we enable our customers to achieve substantially higher6525total system performance and more favorable uptime relative to other AI6526offerings within existing infrastructure cloud environments and unlock6527speed at scale.6528    Policies that ensure a level playing field for all industry6529stakeholders help ensure that technical merit and innovation determine6530success.6531                                 ______65326533    Response to Written Questions Submitted by Hon. Brian Schatz to6534                            Michael Intrator6535Energy Consumption and Cost of Winning6536    Question 1. Do you support President Trump's efforts to expand coal6537power for AI data centers?6538    Answer. At CoreWeave, we believe that there are abundant solutions6539for powering AI data centers. Our company is committed to advancing AI6540infrastructure powered by modern, efficient energy technologies. While6541we understand the challenges of meeting the rapidly growing electricity6542demands of AI data centers, we believe the future lies in a diversified6543energy mix that prioritizes renewables, nuclear power, and fossil6544energy sources using the most efficient, low emitting technologies.65456546    Question 2. How are you addressing the costs associated with new6547infrastructure development in the short-term, including in terms of6548water consumption, pollution, and climate impacts?6549    Answer. CoreWeave prioritizes balancing growth with also being6550efficient and responsible with resources. Regarding water consumption,6551CoreWeave often deploys closed-loop, air-cooled infrastructure and6552liquid cooling technologies that drastically reduce or eliminate water6553use compared to traditional data centers. Many of the data centers6554where we operate provide non and low-emitting energy sources for our6555operations.65566557    Question 3. What plans do you have to source clean energy and to6558publicly report your companies' emissions, if you are not already?6559    Answer. CoreWeave is currently in several data centers that use6560clean energy, have on-site renewables and/or buy Renewable Energy6561Certificates. As we expand into new data centers, we are actively6562engaged in procuring clean energy to power our services.6563    To meet its compliance obligations under California law SB 253,6564CoreWeave will report its Scope 1 and Scope 2 greenhouse gas emissions6565in 2026.65666567    Question 4. Do you believe current incentive structures drive or6568ignore the opportunity for energy innovation?6569    Answer. Past and current incentive structures have helped to drive6570down the cost of all energy sources during the 21st century. These6571include research and development and tax policies.6572    To maintain and strengthen the U.S. position in the AI-continued6573policies are required to ensure adequate supplies of clean, reliable6574supplies of affordable energy CoreWeave supports policies which will6575enable a balanced diverse portfolio of energy technologies including6576advanced nuclear, carbon capture and sequestration, and continued6577advances in renewable energy. The nation also needs to modernize the6578grid and improve energy efficiency.6579PRC Deployment of AI6580    Question 5. What do you believe is the greatest national security6581threat posed by the People's Republic of China's deployment of AI6582systems?6583    Answer. The national security concerns surrounding the People's6584Republic of China's (PRC) deployment of AI systems are interconnected.6585These include the integration of AI into their long-standing civil-6586military fusion strategy, resulting in dual-use capabilities that6587further their economic and defense goals. At the same time, the PRC6588aims to expand its global AI influence by establishing its AI ecosystem6589as a foundational element in numerous countries, increasing its6590capacity to shape the rules governing AI development and deployment6591worldwide.65926593    Question 6. What are your recommendations for addressing these6594threats?6595    Answer. To maintain its leadership in AI and effectively address6596national security threats, the U.S. must scale its AI ecosystem and6597foster innovation across all layers of the AI stack. Carefully6598calibrating export controls and trade agreements to address security6599concerns will limit the PRC's access to vital AI technologies and6600diminish its global AI influence. Enabling the diffusion of the6601American AI ecosystem to allies and like-minded nations that commit to6602security and technology frameworks will also address this threat.6603Strengthening American AI leadership will ensure that democratic values6604and security interests shape the rules governing global AI development.6605Accelerating Scientific Research6606    Question 7. What barriers do you see to making advanced cloud6607computing capacity available to academic and national lab researchers?6608    Answer. We are well-positioned to serve academic and national lab6609researchers because we offer products designed with those users in6610mind. We may face challenges in finding available capacity due to6611increasingly large demand from the private sector and AI labs.66126613    Question 7a. How could Congress help to responsibly reduce those6614barriers?6615    Answer. Congress could consider establishing opportunities and6616resources to enable partnerships between the private sector and Federal6617research centers. A dedicated resource or consortium that aggregates6618computing demand across research institutions could create more6619procurement power and reduce uncertainty that currently acts as a6620constraint.6621    Relatedly, implementing streamlined, standardized procurement6622vehicles for advanced cloud computing could reduce administrative6623barriers that better align with both research timelines and industry6624business models.66256626    Question 8. Should there be dedicated Federal channels or6627procurement pathways to ensure U.S. Federal researchers can access AI6628compute at the pace of scientific need?6629    Answer. Compute allocation is presently dictated through large6630procurement contracts with private sector companies. We seek out and6631prioritize these contracts as they help us derisk our financial6632operations due to their large size and length. This leads us to6633deprioritize shorter-term and smaller-value contracts. The Federal6634government could consider mechanisms to aggregate demand across6635academic and national lab institutions to ensure they can access the6636compute needed to maintain the U.S. leadership in AI.6637                                 ______66386639   Response to Written Questions Submitted by Hon. Edward Markey to6640                            Michael Intrator6641Comprehensive Impacts of Data Center Construction6642    Question 1. When planning for data center construction, does your6643company conduct a cradle-to-grave infrastructure study that includes6644wildlife, community, and pollution impacts during and beyond the6645operational lifespan of a data center? If yes, what have you learned6646from those studies? If no, why not?6647    Answer. CoreWeave is developing a more formal site selection6648process that will ensure we are tracking climate risk and resiliency,6649as well as examine broader ecosystem risks. While we do not currently6650conduct formal cradle-to-grave lifecycle analyses for every facility,6651our approach integrates rigorous environmental due diligence,6652stakeholder engagement, and adaptive best practices to minimize6653ecological and community impacts.6654    We select sites with adaptive reuse as the focus and we often6655retrofit existing structures rather than developing greenfield sites,6656allowing us to have less impact on land use. In addition, we focus on6657using our existing data centers as efficiently as possible so that we6658can fit more servers into less space, therefore reducing the need for6659overall square footage.6660Backup Energy Generation6661    Question 2. Does your company use backup diesel generators at any6662facilities?6663    Answer. Our data center providers use backup diesel generators in6664the data centers in which we operate. These backup generators are6665assets that belong to our data center providers. Their service to us6666includes redundancy to the utility power source for the data center.66676668    Question 2a. If yes, please provide a list of each facility where6669diesel generators are being used, along with the location, quantity,6670and type of generators.6671    Answer. Our data center providers use backup generators, and these6672assets belong to the providers.66736674    Question 2b. If yes, did your company consider the use of battery6675storage technology as an alternative to diesel generators? Please6676explain your decision process.6677    Answer. We do this currently. Our data centers have uninterruptible6678power supply infrastructure in place that would cover the immediate 4-66679minutes of a power outage to the data center.6680    CoreWeave is exploring ways to incorporate battery storage into our6681back-up generation as part of maintaining our competitive edge with a6682focus on network resiliency. Backup battery storage would allow us to6683have greater resiliency for CoreWeave's operations and the grid6684overall.6685    We see battery storage as a complementary technology that can6686further strengthen our microgrid capabilities, manage renewable6687intermittency, and provide fast-responding backup for mission-critical6688workloads.6689Energy Mix6690    Question 3. Does your company utilize any on-site or colocated6691energy generation to power your data centers?6692    Answer. No.66936694    Question 3a. If yes, please provide detail how much power comes6695from on-site and colocated energy generation.66966697    Question 3b. If yes, please list all on-site and colocated energy6698sources (e.g., renewable, nuclear, hydropower, gas-powered turbines,6699etc.) that are being utilized to power your data centers.67006701    Question 4. How does your company ensure local ratepayers are not6702responsible for paying the cost of new energy infrastructure, such as6703transmission lines, needed to meet the data center's energy demand?6704    Answer. As we review new sites, we are assessing the cost structure6705of our energy. For at least one of our large expansions coming later6706this year, the rate CoreWeave pays will include additional surcharges6707in order to fund rate protection for local ratepayers. We understand6708that there is a concern that incumbent ratepayers will bear the cost of6709increased data center power demand. CoreWeave is committed to paying6710its fair share of the costs required to meet its power demand.6711Energy Consumption in AI Model Training6712    Question 5. Data centers require vast amounts of water for cooling.6713When water is at critically low levels, does your company continue to6714pull water for building cooling? Does it have a contingency for6715operating as to not put further stress on the water supply and6716potentially take limited resources from households?6717    Answer. CoreWeave aims to reduce the amount of water it uses and we6718have a different approach to cooling data centers. Whenever possible,6719we set up air cooled chillers, which are a closed loop system. Once the6720system is filled, there is a limited amount of water required. Each6721data center is evaluated to consider the environmental and climate6722cost-benefit analysis of power usage versus water usage.6723                                 ______67246725    Response to Written Questions Submitted by Hon. Jacky Rosen to6726                            Michael Intrator6727Competition Across the AI Ecosystem6728    It's essential that as we build a strong AI industry in the U.S.,6729we also focus on establishing a competitive one.67306731    Question 1. What should Congress consider to promote competition in6732the AI ecosystem and across the tech stack, such as interoperability6733requirements or securing access to computing power for researchers?6734    Answer. Competitive dynamism is essential for continued American AI6735leadership. Policies that ensure a level playing field for all industry6736stakeholders will benefit American dynamism and innovation more6737broadly. In this vein, it is important to streamline government6738certification processes, ensure the efficient administration of6739existing requirements like export license reviews, and ensure that6740newer entrants can effectively bid for government contracts. The key is6741ensuring that technical merit and innovation determine success.6742    In addition, open standards are an important factor for operating6743our cloud at scale. CoreWeave currently participates in industry groups6744that define open standards and hardware interoperability. Without these6745industry groups defining hardware interoperability, CoreWeave would not6746be able to deploy the heterogeneous mix of hardware that our customers6747demand at the scale and speed that they need. Open standards can6748enhance supply chain resiliency and accelerate innovation, while also6749reducing barriers for different and new stakeholders to participate.6750Model Security6751    Question 2. It's essential we ensure the AI models we use do not6752become another cybersecurity vulnerability. Would voluntary6753cybersecurity standards for large AI models or high-risk models and the6754infrastructure they were trained on be helpful in establishing trust?6755    Answer. Yes, voluntary cybersecurity standards for AI6756infrastructure can be helpful in establishing trust and establishing6757common vocabulary for stakeholders, particularly through industry-led6758initiatives.6759                                 ______67606761    Response to Written Question Submitted by Hon. Roger Wicker to6762                               Brad Smith6763Fiber Optic Cable Supply Chain6764    Background: Despite some domestic manufacturing expansion, the U.S.6765remains heavily dependent on foreign imports of fiber optic cable from6766Thailand, with a particular dependency on components from China. In67672024, China had 300 million km of excess fiber capacity, which has6768depressed global prices.67696770    Question. Mr. Smith, when we talk about AI infrastructure, we often6771focus on compute needs and the gaps in sufficient energy to power AI6772data centers. But there's also a critical need for infrastructure. This6773committee often considers fiber infrastructure in terms of its6774importance to broadband, but it is also essential for AI and our6775ability to stay ahead in AI development. Can you describe the6776importance of connecting AI data centers to one another and why this6777matters?6778    Answer. AI workloads require significant computational power and6779fiber networks play a crucial role in delivering high-speed, low-6780latency connections for real-time data transfer in and out of6781datacenters. Hyperscalers often have significant fiber optic transport6782infrastructure, some of which they own and some of which they contract6783with telecom service providers to use. For example, Microsoft has6784constructed a fiber-optic AI wide-area network connecting its data6785center footprint with a 400 terabyte per second fiber-optic backbone,6786capacity that is ten times what was enabled for traditional data6787centers. This private network is critical to the functioning of the6788Microsoft cloud as it moves data between datacenters.6789    Connecting AI data centers to one another is vital for enhancing6790performance, reliability, and scalability for AI operations.6791Interconnected centers enable distributed processing and optimize6792resource utilization while facilitating load balancing to ensure no6793single datacenter is overwhelmed. Enabling redundancy and failover6794mechanisms also enhances disaster recovery and ensures continuous6795operation. Public networks play a crucial role too and must be6796sufficiently robust to reliably carry data to and from the data center6797network.6798                                 ______67996800     Response to Written Questions Submitted by Hon. Todd Young to6801                               Brad Smith6802Federal Investment in Science Research6803    America's global leadership in technology didn't happen by6804accident--it's been built on decades of strong, sustained Federal6805investment in basic research.68066807    Question 1. Mr. Smith, Can you speak to why that kind of6808foundational support from the Federal government remains essential6809today, and how it has helped position the United States as a global6810engine of innovation, particularly in emerging fields like AI and6811quantum?6812    Answer. For the last 80 years, the United States has led the world6813with its scientific and technological prowess, resulting in6814transformative products and capabilities. To outcompete nations like6815China, we must significantly boost our R&D investments and ensure we6816have skilled researchers and scientists focused on emerging6817technologies. Experts predict China will continue to invest substantial6818resources into next-generation technologies such as AI, advanced6819manufacturing, clean energy, quantum computing, and semiconductors over6820the next decade. The United States needs the same level of6821intentionality today that we had in the early 20th century--where a6822deliberate effort brought together industry, government and academia to6823propel scientific advancement.6824    AI technology is at an inflection point. It is precisely at this6825stage--when scientific breakthroughs are on the cusp of scaling--that6826public investment matters most. Cuts to basic research, particularly in6827AI, risk delaying or even derailing the U.S. trajectory in this modern6828AI moment. Without predictable, long-term investment, the United States6829will fall behind in both scientific leadership and the6830commercialization of critical technologies.6831AI Public Awareness and Education6832    Winning the diffusion race not only requires providing a pathway6833for greater adoption of the technology and its applications into the6834general stream of commerce but also bolstering our public awareness and6835education of AI.68366837    Question 2. Mr. Smith, can you speak to how the Secretary of6838Commerce can have a whole of government approach to fostering greater6839awareness for AI literacy and growing STEM opportunities to create the6840next generation of our workforce?6841    Answer. AI, like all new technologies, will disrupt the economy and6842displace some jobs. However, we believe AI will help lower the barriers6843to entry for many professions, replace rote tasks, and create a6844foundation for human creativity that builds on AI tools. AI will create6845new economic opportunities, allowing entrepreneurs to start new6846businesses and create new jobs. We are already seeing some of these6847benefits both at Microsoft and across the economy. A recent LinkedIn6848report highlighted research on how businesses using AI are seeing the6849benefits in innovation and creativity, and even in expanding their6850workforce.6851    Americans of all ages and backgrounds need AI skills to complete in6852this new world of work. A key opportunity for most people will be to6853develop an AI fluency that will enable them to use AI in their jobs,6854much as they use laptops, smartphones, software applications, and the6855Internet today.6856    The U.S. Department of Commerce includes key agencies like NIST,6857NTIA, MBDA, and the Census Bureau, each playing a unique role in6858economic development, data governance, and innovation. AI literacy can6859be integrated across these departments by aligning with their6860missions--for example, NIST can support standards-based AI education,6861and NTIA can embed AI training in their digital programs. Recent6862legislation and national strategies also call on the Department to lead6863public awareness campaigns and fund AI learning initiatives through6864libraries, nonprofits, and workforce programs.6865AI and Quantum6866    We know AI and quantum computing are both strategic frontiers, but6867what's becoming clearer is that the breakthroughs we'll need most may6868come from how these technologies interact.68696870    Question 3. Mr. Smith, from your vantage point, how should the6871United States be thinking about the convergence of AI and quantum--not6872just as two separate priorities, but as part of a unified strategy to6873outpace China in foundational technologies?6874    Answer. Artificial Intelligence and quantum computing are two6875strategic frontiers of technology--and their convergence is poised to6876unlock unprecedented breakthroughs. In our view, both AI and quantum6877computing are ``foundational technologies'' that will drive innovation6878across industries, from healthcare to defense. We believe there is a6879powerful synergy in combining AI with quantum computing. For example,6880quantum computing can vastly accelerate AI. Certain computations that6881choke classical computers can run exponentially faster on quantum6882machines. This hints at quantum's potential to turbocharge AI tasks6883like machine learning optimization, large-scale data analysis, and6884complex simulations. In practical terms, a future quantum co-processor6885might, for example, dramatically speed up a clustering algorithm6886working on high-dimensional data or enable AI to analyze combinatorial6887scenarios that are infeasible for classical computers. Conversely, AI6888can help quantum progress. AI algorithms assist in designing better6889quantum circuits and error-correction methods, essentially using AI to6890overcome quantum engineering challenges. AI can also generate synthetic6891data or heuristics to guide quantum algorithm development--essentially,6892using AI to explore which problems a quantum computer should tackle and6893how. By co-developing these technologies, we unlock new capabilities--6894from more accurate drug discovery (e.g., AI identifying candidates,6895quantum evaluating molecular interactions) to climate modeling and6896materials science breakthroughs.6897    Harnessing this dual momentum is not just a scientific imperative,6898it is a strategic necessity. Given China's coordinated, state-led6899advancements in both artificial intelligence and quantum technologies,6900we believe the United States must pursue bold, comprehensive investment6901across all dimensions: research, policy, industry, and workforce. This6902includes targeted support for their convergence, cross-disciplinary6903collaboration, and enabling private-sector innovation.6904                                 ______69056906   Response to Written Questions Submitted by Hon. Maria Cantwell to6907                               Brad Smith6908AI Standards6909    The U.S. driving development of AI standards alongside the most6910advanced democracies in the world offers us an opportunity to set the6911``rules of the road'' for AI on the global stage.69126913    Question 1. How do NIST standards help the United States'6914competitiveness?6915    Answer. Industry-led voluntary standards are instrumental in6916driving U.S. competitiveness by establishing globally recognized6917benchmarks for quality, safety, and interoperability in AI systems.6918These standards ensure that American AI technologies are seen as6919reliable, trustworthy, and high performing, which is critical for6920adoption both domestically and internationally. NIST's convening6921function across government, industry, and civil society encourages6922active participation by U.S. entities in the development of pre-6923standardization materials--which will ultimately influence the global6924AI landscape. NIST's priorities--as a reflection of the6925Administration's priorities--also provide an important signal to6926industry, encouraging strategic alignment on shared values and global6927trade.69286929    Question 2. What standards would you like to see NIST develop or6930promote to improve U.S. competitiveness?6931    Answer. Today, there is a dearth of consensus reference points--6932even across industry--that identify the most likely AI risks and how to6933evaluate and mitigate them reliably. Microsoft is working with industry6934peers to develop consensus best practices, but there are opportunities6935to expand and accelerate these efforts. NIST's expertise in test,6936evaluation, validation, and verification (TEVV) and measurement science6937should be leveraged to provide structure and guidance to developers and6938deployers on how to best develop and use evaluations to accelerate6939adoption while minimizing risks. For example, NIST could develop6940practices for evaluating evaluations to ensure they are scientifically6941valid and could assemble panels of evaluations for use by the6942government and industry to evaluate specific risks or capabilities.6943AI Exports and Export Controls6944    The U.S. needs a strong national export strategy for technology and6945other U.S. exports. Alliances create markets and engaging our allies is6946essential to effective coordination and implementation of export6947controls.69486949    Question 3. What is the best way for the U.S. Government--6950specifically, the U.S. Department of Commerce--to support U.S.6951companies in exporting AI technologies? Where should we target to make6952sure our foreign adversaries do not get there first?6953    Answer. Congress and the Administration both have key roles in6954protecting our national security by preventing adversaries from6955obtaining advanced technology components. Policies must be balanced so6956that American companies can thrive and set the global standard for the6957technology.6958    The proposed ``AI Diffusion'' rule, sets forth a robust set of6959security standards expected of companies building AI datacenters. We6960believe these qualitative guardrails point in a sensible direction for6961U.S. policy especially as it relates to national security concerns.6962However, we disagree with the ``tiering'' of countries as well as6963certain quantitative restricts--absent a universal license--as we think6964arbitrary restrictions undermine international confidence in access to6965critical American AI technology and therefore restrict the ability of6966U.S. firms to compete globally.69676968    Question 4. You mentioned that countries around the world will only6969use American AI if they can trust it. How can the U.S. government6970partner with the private sector to build that trust?6971    Answer. Business planning and investment decisions rely heavily on6972predictable and trusted relationships on both sides. By the same token,6973uncertainty complicates and slows decision making and execution. Clear6974and consistent policies enable companies like Microsoft to make long-6975term commitments to innovation, workforce development, and6976infrastructure investment both abroad and domestically. The Federal6977government has an opportunity to lead in their own procurement rules6978and the adoption and utilization of AI products and services.6979Privacy6980    Data is a foundational element in the tech stack for any AI system.6981Advances in AI will spur an increase in demand for data, both to train6982and ground AI models. This enhances the need for bright lines related6983to consumer data collection and usage. The best way to set these bright6984lines is through a strong, comprehensive Federal data privacy law.69856986    Question 5. Do you agree that bright lines around consumer privacy6987will spur innovation in artificial intelligence?6988    Answer. Yes, raising privacy protections will spur innovation and6989benefit the U.S. economy by bolstering consumer confidence and ensuring6990that data will be appropriately protected while still being available6991for beneficial uses. Both components are critical to AI development.6992Consumers remain deeply concerned about the collection and processing6993of their personal information.\1\ They report confusion about where it6994is going and cite concern over a loss of privacy from social media, AI,6995and impact on children.\2\ They also report little faith in6996government's ability to solve the problem, with 71 percent saying that6997they do not expect social media companies to be reined in even as 776998percent say they would like more government action.\3\ Concern is6999shared across the political spectrum with 68 percent of Republicans and700078 percent of Democrats saying there should be more regulation over7001personal information.\4\ By some measures the problem is actually7002getting worse.7003---------------------------------------------------------------------------7004    \1\ How Americans View Data Privacy, Colleen McClain, Michelle7005Faverio, Monica Anderson and Eugenie Park, Pew Research Center, October700618, 2023. See: https://www.pewresearch.org/internet/2023/10/18/how-7007americans-view-data-privacy/7008    \2\ Id.7009    \3\ Id.7010    \4\ Id.7011---------------------------------------------------------------------------7012    The share of adults reporting that they understand privacy laws7013very little/not at all has risen from 63 percent in 2019 to 72 percent7014in 2023.\5\ None of this should be a surprise. Consumers are deluged by7015the ways that lost or misused personal information can harm them.7016Whether it's through data breach notifications and fears over7017increasingly sophisticated identity theft scams,\6\ misuse of personal7018information by employers,\7\ or unexpected negative outcomes such as a7019fitness app revealing the location of secret U.S. military bases,\8\7020the use and misuse of personal information can spill into every facet7021of a consumer's life.7022---------------------------------------------------------------------------7023    \5\ Id.7024    \6\ Verizon 2024 Data Breach Investigations Report, see: 2024 Data7025Breach Investigations Report | Verizon7026    \7\ Michelle Boorstein, Marisa Iati, and Annys Shin, ``Top U.S.7027Catholic Church official resigns after cellphone data used to track him7028on Grindr and to gay bars,'' Wash. Post (July 21, 2021), see: https://7029www.washingtonpost.com/religion/2021/07/20/bishop-misconduct-resign-7030burrill/ .7031    \8\ Jeremy Hsu, ``The Strava Heat Map and the End of Secrets'',7032Wired (Jan 29, 2018), see: Strava Data Heat Maps Expose Military Base7033Locations Around the World | WIRED7034---------------------------------------------------------------------------7035    This lack of consumer confidence is directly impacting economic7036growth. A recent report by the World Trade Organization and the OECD7037found that if all nations adopted privacy safeguards, global exports7038would increase by 3.6 percent and global GDP would grow by 1.777039percent.\9\ An absence of privacy protections would cause global GDP to7040fall by almost 1 percent and global imports by over 2 percent with the7041largest impact on high-income economies.\10\ A fragmented U.S. privacy7042landscape--such as the state patchwork currently in place in the US--7043also risks negatively impacting the U.S. economy. One industry estimate7044puts the cost of a patchwork of state laws at over $1 trillion over 107045years, with $200 billion of that burden falling on small7046businesses.\11\ Given these important benefits and looming potential7047costs, passage of Federal comprehensive privacy legislation is critical7048for spurring AI innovation.7049---------------------------------------------------------------------------7050    \9\ OECD/WTO (2025), Economic Implications of Data Regulation:7051Balancing Openness and Trust, OECD Publishing, Paris, https://doi.org/705210.1787/aa285504-en.7053    \10\ Id pg. 39.7054    \11\ ITIF, ``The Looming Cost of a Patchwork of State Privacy7055Laws,'' (January 2022) see: https://itif.org/publications/2022/01/24/705650-state-patchwork-privacy-laws-could-cost-1-trillion-more-single-7057federal/70587059    Question 6. Will Microsoft work with me on Federal comprehensive7060privacy legislation to set those bright lines?7061    Answer. Yes, Microsoft will work with you and any other members of7062congress on Federal comprehensive privacy legislation. We've been7063advocating for it since 2005, because we believe it is critical to7064building consumer trust and driving economic growth in the United7065States.\12\7066---------------------------------------------------------------------------7067    \12\ ``Microsoft Advocates Comprehensive Federal Privacy7068Legislation,'' Microsoft News, November 3, 2005, see: https://7069news.microsoft.com/2005/11/03/microsoft-advocates-comprehensive-7070federal-privacy-legislation/.7071---------------------------------------------------------------------------7072Energy Needs and R&D for Fusion Energy7073    The growing demand for electricity to power AI data centers is7074staggering. By some estimates, global electricity demand from data7075centers is projected to more than double by 2030 exceeding 9457076terawatt-hours (TWh). It will strain electric grids and energy7077providers.70787079    Question 7. What plan does your company have to meet energy needs7080for AI, and what investments are you making into non-fossil fuel7081sources of energy such as fusion?7082    Answer. As we consider any power sources, we look for reliable,7083scalable and cost-effective options that can be developed in rapid7084timelines aligned to our datacenter growth. To match our carbon7085emitting electricity use, Microsoft has entered into power purchase7086agreements to add 34 GW of renewable energy capacity globally. In 2024,7087we entered into a power purchase agreement to add 835 MW of carbon-7088free, nuclear energy from Constellation's Crane Clean Energy Center7089that is anticipated to be put into service by 2028, which will provide7090additional capacity to the PJM interconnect.7091    In 2023, we announced an agreement with Helion Energy to procure7092power from its first fusion power plant that is under development. We7093are optimistic that fusion energy can play an important part of the7094grid mix in the future.70957096    Question 8. With respect to fusion energy, how can the government7097partner with the private sector to scale fusion technology as it7098continues to develop?7099    Answer. To achieve fusion breakthrough and scale fusion energy,7100government and private sector partnerships are essential in order to7101reduce risk, accelerate innovation, and build the foundation for fusion7102commercialization. The government can support cost-sharing programs and7103adopt clear, adaptive regulatory frameworks specific to fusion to7104accelerate deployment. Aligning public resources with private7105innovation can help bring fusion energy to market faster and more7106affordably.7107AI Safety7108    We are seeing a proliferation of deepfakes and other AI content7109that threatens the average person's ability to discern truth in media.7110And that's just one area in the field of AI that presents complicated7111safety questions. The U.S. AI Safety Institute plays a critical role in7112ensuring that AI systems are developed responsibly and that the most7113advanced models are fully tested. This is crucial for building trust7114and promoting wider adoption.71157116    Question 9. Do you support the work of the U.S. AI Safety7117Institute?7118    Answer. Yes, we support the work of the U.S. Center for AI7119Standards and Innovation, formerly known as the U.S. AI Safety7120Institute. Their efforts to promote reliable development of artificial7121intelligence, establish rigorous safety standards, foster cross-sector7122collaboration, and address potential risks are crucial for responsible7123innovation. We share their commitment to ensuring AI benefits society7124through principles of reliability, transparency, and accountability.7125AI Workforce7126    The U.S. needs a skilled workforce to build and maintain AI7127infrastructure, including electricians, pipefitters, carpenters, and7128engineers. Labor shortages are already slowing down data center7129construction, yet there is no national strategy to train or retain this7130talent.71317132    Question 10. What do you think Congress and the administration7133should be doing to support AI education, training, and workforce7134development?7135    Answer. AI, like all new technologies, will disrupt the economy and7136displace some jobs, and we which we know causes concern for many people7137and the workforce at large. We believe AI will create new opportunities7138that will outweigh many of the challenges ahead. AI will help lower the7139barriers to entry for many professions, replace rote tasks, and create7140a foundation for human creativity that builds on AI tools. AI will7141create new economic opportunities, allowing entrepreneurs to start new7142businesses and create new jobs. We are already seeing some of these7143benefits both at Microsoft and across the economy. A recent LinkedIn7144report highlighted research on how businesses using AI are seeing the7145benefits in innovation and creativity, and even in expanding their7146workforce. We encourage congress to invest in Federal apprenticeships7147focused on major AI infrastructure initiatives, establish Federal7148programs for on-the-job training and support the reauthorization of the7149Workforce Innovation and Opportunity Act.71507151    Question 11. What challenges do your companies and others face from7152the administration's immigration policies, and what concerns do you7153have about impacts on high-skilled immigration?7154    Answer. The United States faces a critical talent bottleneck. Our7155universities educate some of the world's most talented engineers,7156scientists, and entrepreneurs, but outdated immigration policies often7157force them to leave the country shortly after graduation. Companies7158like ours face long delays and constant uncertainty in the visa and7159green card process, which makes it difficult to hire and retain the7160talent we need. In high-demand fields like artificial intelligence,7161advanced engineering, and quantum computing, the need for specialized7162expertise far exceeds the available domestic supply.7163    This isn't about replacing American workers. It's about enabling7164innovation by complementing the U.S. workforce with global talent. To7165build the technologies of the future, we need to work with the best7166minds from around the world. American companies serve customers across7167the globe, and having a smart people from around the world work on our7168teams helps us create products that truly meet the needs of a global7169audience.7170    To remain globally competitive, the United States needs an7171immigration system that attracts and retains top talent. That means7172creating fast, reliable pathways for highly skilled individuals--7173especially graduates of U.S. universities--and addressing the green7174card backlog for workers from countries facing extreme wait times. If7175we want to lead the world in AI and other emerging technologies, we7176must continue to be a magnet for the world's best and brightest--and7177ensure they have the opportunity to build a future in the United7178States.7179                                 ______71807181    Response to Written Questions Submitted by Hon. Brian Schatz to7182                               Brad Smith7183Future of Work7184    Question 1. What is your vision of the future of work and what are7185the valuable jobs of the future, in the near-term and long-term?7186    Answer. We believe that AI is going to impact the future of work7187tremendously--largely for the better. Additionally, this is an7188important moment for the government to be engaged in these7189conversations. At the beginning of the year, I published a blog post7190connected to this question. LinkedIn also recently published a report,7191AI and the Global Economy, examining how AI is already impacting the7192workforce and the economy at large.71937194    Question 2. How is your company taking advantage of the automation7195you're empowering to scale productivity without leaving workers behind?7196    Answer. We design our technologies to augment human capability--7197empowering people to achieve greater impact with each hour they spend7198at work. That means investing in broad-based skilling initiatives,7199including apprenticeships and workforce partnerships, so every worker,7200not just those in tech, can benefit from AI. We encourage employees and7201partners to experiment with tools like Copilot to find the best use7202cases for their work. And we've built feedback mechanisms to ensure7203continuous improvement and keep worker experience at the center of our7204innovation.7205PRC Deployment of AI7206    Question 3. What do you believe is the greatest national security7207threat posed by the People's Republic of China's deployment of AI7208systems?7209    Answer. The People's Republic of China's deployment of AI systems7210presents a range of national security risks. For example, the spread of7211Chinese technology to third countries may undermine global7212cybersecurity and information integrity. Through advanced AI-driven7213surveillance, cyber-espionage, and disinformation campaigns, China7214could exploit vulnerabilities in critical infrastructure, manipulate7215public opinion, and erode trust in democratic institutions.72167217    Question 4. What are your recommendations for addressing these7218threats?7219    Answer. Microsoft advocates a comprehensive, multi-layered approach7220to counter the national security risks posed by China. To address AI-7221driven surveillance and cyber-espionage, Microsoft recommends deploying7222zero-trust architectures, enhancing endpoint detection and response7223capabilities, and investing in AI-powered threat intelligence to detect7224and mitigate advanced persistent threats. Strengthening public-private7225partnerships is essential to ensure real-time information sharing and7226coordinated responses to cyber intrusions.7227    Microsoft also recognizes the risks posed by nation-state actors7228including China employing cyber enabled influence operations targeting7229critical institutions within America. Microsoft harnessed the data7230science and technical capabilities of our AI for Good Lab and Microsoft7231Threat Analysis Center (MTAC) teams to assess these risks including7232whether or not these actors were utilizing AI in these operations. When7233appropriate, the team calls on the expertise of Microsoft's Digital7234Crimes Unit to invest in and operationalize the early detection of AI-7235powered criminal activity and respond fittingly, through the filing of7236affirmative civil actions to disrupt and deter that activity and7237through threat intelligence programs and data sharing.7238    In addition, Microsoft is committed to advancing information7239integrity and believes that including content credentials is an7240important driver for this. We are a founding member of the Coalition7241for Content Provenance and Authenticity (C2PA). To achieve7242transparency, support information integrity, and empower our users, we7243are leveraging C2PA's ``content credentials'' open standard across7244several products. For example, content containing the ``Content7245Integrity'' technology has been automatically labelled on LinkedIn,7246with users beginning to see the ``Cr'' icon on images and videos that7247contain C2PA metadata.7248    Beyond technological solutions that improve defenses, Microsoft7249also stresses the importance of political solutions to these threats--7250governments taking a more proactive and coordinated role in7251attributing, exposing, and deterring Chinese malicious cyber activity.7252This includes timely public attribution, diplomatic pressure, and legal7253action where appropriate. Clear consequences for state-sponsored7254cyberattacks are essential to shift the cost-benefit calculus and7255reinforce international norms against digital aggression.7256    This is especially true for cyber intrusions which target critical7257infrastructure in order to ``preposition'' for disruptive or7258destructive attacks in a future contingency. Such operations put7259civilians at significant risk and are meaningfully different from7260traditional espionage. Prepositioning cyberattacks should be recognized7261as a ``threat'' of force prohibited by international law that must be7262deterred via sufficient consequences across domains.7263    Finally, ensuring that U.S. AI technology retains its global7264leadership and that regulations and policies do not unnecessarily7265hinder the global diffusion of U.S. AI is a key component to countering7266China.7267Accelerating Scientific Research7268    Question 5. Microsoft's AI for Health and AI for Earth programs7269have supported hundreds of academic and nonprofit research projects.7270How would you suggest the government can better structure7271collaborations to accelerate scientific discovery using AI?7272    Answer. *See answer below.72737274    Question 6. You've led efforts to aggregate and standardize7275environmental and biomedical datasets. What further steps should7276industry and government take together to ensure researchers have access7277to well-structured, interoperable, and annotated datasets for AI-driven7278discovery?7279    Answer. The government can play a pivotal role in increasing access7280to datasets and accelerating scientific discovery by supporting7281initiatives that make data more discoverable, accessible, and usable.7282Launching a national open data campaign would empower institutions such7283as the National Archives, the Library of Congress, the Smithsonian, and7284other government agencies to digitize, organize, and share non-7285classified and non-sensitive data for AI training. Additionally, the7286creation and expansion of open data commons within non-profit,7287academic, and cultural institutions could further democratize access to7288valuable datasets. Addressing paywalls that restrict access to7289scientific research and establishing dedicated funds to unlock closed-7290access journals would also contribute significantly to the availability7291of critical knowledge for innovation.7292    The government should also prioritize the development and7293enforcement of clear guidelines for data sharing and annotation.7294Stronger incentives and monitoring mechanisms are needed to ensure7295timely and secure sharing of federally funded research data, in7296alignment with funding agreements. Supporting the adoption of metadata7297and provenance standards for datasets can enhance their utility and7298reliability, particularly in the era of synthetic data generated by AI7299models. By taking these steps, the government can foster a robust7300ecosystem where data serves as the backbone for scientific advancement7301and technological progress.7302Energy Consumption and Cost of Winning7303    Question 7. Do you support President Trump's efforts to expand coal7304power for AI data centers?7305    Answer. The availability of reliable, resilient, and cost-effective7306electricity is essential for economic growth in the United States. The7307recent Executive Orders from the President on coal generation highlight7308the urgency to meet energy sector demand growth, including as a bridge7309to future carbon free generation as new energy options are built out.7310At Microsoft, when we consider any potential dedicated power source, we7311look for generation options that align with our need for reliable,7312cost-effective, and sustainable electricity. We also invest in the next7313generation of energy supply technologies to support accelerated7314innovation and cost-declines. In practice, this has meant that we use7315and invest in a diverse mix of electricity generation technologies that7316align with our needs for reliable and cost-effective electricity. These7317technologies also must align with our commitment to be carbon negative7318by 2030. Across all of these priorities--reliability, cost-7319effectiveness, and sustainability--we do not see a business case for7320expanded use of coal to meet the energy needs of our data centers.7321Rather, we see increasingly strong arguments for integrating more7322renewables, nuclear, net-zero, and net-negative electricity generation7323technologies into the system in addition to a diverse set of energy7324storage technologies. While many of these new technologies present7325promising options in the future, we also recognize that there will7326sometimes be a need to build other options in the near term as we grow7327our Nation's energy infrastructure.73287329    Question 8. How are you addressing the costs associated with new7330infrastructure development in the short-term, including in terms of7331water consumption, pollution, and climate impacts?7332    Answer. When improvements to the grid are required to serve our7333load, we work with our local utility to ensure that we pay for the7334improvements required to serve our site. We are thoughtful about the7335resource intensity of AI from the moment that we decide to build a7336datacenter. We're working to advance low-carbon materials and create7337global markets to help advance sustainability across industries. When7338operating our datacenters, we optimize energy and water efficiency,7339including announcing a new datacenter design that consumes zero water7340for cooling. We also are increasingly adopting a circular approach to7341reach our target of zero waste by 2030. We target preventing waste7342first, then focus on reusing, and recovering materials. This includes7343reusing and recycling construction and demolition waste, diverting7344operational waste and advancing circular cloud hardware and packaging.7345                                 ______73467347   Response to Written Questions Submitted by Hon. Edward Markey to7348                               Brad Smith7349Comprehensive Impacts of Data Center Construction7350    Question 1. When planning for data center construction, does your7351company conduct a cradle-to-grave infrastructure study that includes7352wildlife, community, and pollution impacts during and beyond the7353operational lifespan of a data center? If yes, what have you learned7354from those studies? If no, why not?7355    Answer. Datacenters are long-term investments, and planning is a7356multi-year, capital intensive program that requires alignment across a7357range of factors to be successful. These factors include energy, water,7358fiber, land suitability, environmental considerations including7359wildlife impacts, and an available, trained workforce to ensure that we7360deploy the data center on the timelines our customers expect.7361    We are thoughtful about resource intensity of AI from the moment7362that we decide to build a datacenter. We're working to advance low-7363carbon materials and create global markets to help advance7364sustainability across industries. When operating our datacenters, we7365optimize energy and water efficiency, including announcing a new7366datacenter design that consumes zero water for cooling.7367    We also are increasingly adopting a circular approach to reach our7368target of zero waste by 2030. We target preventing waste first, then7369focus on reusing, and recovering materials. This includes reusing and7370recycling construction and demolition waste, diverting operational7371waste and advancing circular cloud hardware and packaging.7372Backup Energy Generation7373    Question 2. Does your company use backup diesel generators at any7374facilities?7375    Answer. Yes. Generators at datacenters, most often powered by7376diesel fuel, play a key role in delivering reliable backup power so we7377can meet the needs of the many customers that rely on our services,7378including hospitals, first responders, and educational institutions.7379Each of these generators is used for no more than a few hours a year or7380less at our datacenter sites, most often for routine maintenance or for7381backup power during a grid outage.73827383    Question 2a. If yes, please provide a list of each facility where7384diesel generators are being used, along with the location, quantity,7385and type of generators.7386    Answer. We use generators at all of our datacenter sites. A7387complete list of our datacenter sites can be found here. [https://7388datacenters.microsoft.com/globe/explore/] We operate our backup7389generators sparingly.73907391    Question 2b. If yes, did your company consider the use of battery7392storage technology as an alternative to diesel generators? Please7393explain your decision process.7394    Answer. As we consider any back-up power sources, we look for7395reliable, scalable and cost-effective options that can be developed and7396deployed in rapid timelines aligned to our datacenter growth and7397operate within the constraints of our datacenter facilities, including7398batteries and other technologies (e.g., alternative fuels). The primary7399requirements for such backup generation are that they can quickly ramp7400to meet the emergency event and can store sufficient energy to supply7401the datacenter through an outage in case site access is limited.7402Energy Mix7403    Question 3. Does your company utilize any on-site or colocated7404energy generation to power your data centers?7405    Answer. No.74067407    Question 3a. If yes, please provide detail how much power comes7408from on-site and colocated energy generation.7409    Answer. N/A74107411    Question 3b. If yes, please list all on-site and colocated energy7412sources (e.g., renewable, nuclear, hydropower, gas-powered turbines,7413etc.) that are being utilized to power your data centers.7414    Answer. N/A74157416    Question 4. How does your company ensure local ratepayers are not7417responsible for paying the cost of new energy infrastructure, such as7418transmission lines, needed to meet the data center's energy demand?7419    Answer. When improvements to the grid are required to serve our7420load, we work with our local utility to ensure that we pay for the7421improvements required to serve our site.7422Energy Consumption in AI Model Training7423    Question 5. In the past year, how many GWh of energy do you7424estimate was used to train new AI models?7425    Answer. To date, most of our infrastructure sites have been serving7426both AI and traditional cloud services for our customers and the7427critical business and communications requirements they rely on us for.7428AI has shared space and resources in those locations, so we are not yet7429able to specifically separate out AI energy use for prior years. We7430continue to drive efficiency into every part of this infrastructure as7431we deploy this new technology at scale. We are working on further7432analysis of energy use.74337434    Question 6. Data centers require vast amounts of water for cooling.7435When water is at critically low levels, does your company continue to7436pull water for building cooling? Does it have a contingency for7437operating as to not put further stress on the water supply and7438potentially take limited resources from households, agriculture, or7439small businesses?7440    Answer. For the datacenter, water requirements will vary depending7441on the cooling technology used. Microsoft is moving to solutions that7442utilize zero water for cooling in water-stressed areas. However, water7443and sewer connections are typically required for basic safety and7444administrative functions, like restrooms and breakrooms.7445Government Partnerships7446    Question 7. Your company offers AI products specifically for the7447public sector, which are now used across Federal agencies and state and7448local governments. Given the especially heightened risks related to7449governments' use of AI--including the denial of rights or access to7450services and false or incorrect information about government benefits7451and programs--what additional steps have you taken to ensure that these7452tools are safe and effective to use in the context of government?7453    Answer. The Federal government's leadership in AI adoption is vital7454for setting standards and keeping the U.S. at the forefront of7455innovation. By integrating AI into its operations and using existing AI7456applications, it can speed up public service delivery, drive widespread7457adoption, improve services, and boost industry confidence. To do this7458effectively, we've taken additional steps to ensure the tools used in7459government are safe and effective. For example:74607461    Microsoft has cloud instances designed specifically for the public7462sector, meaning specialized versions of its cloud services designed to7463meet unique security, compliance, and operational needs of government7464agencies. For example, Microsoft Azure Government and Microsoft 3657465Government are built to handle sensitive data and regulated workloads7466and are certified to meet the standards of FedRAMP High. While the7467underlying technology is often the same as in our commercial offerings,7468these instances are configured and governed to align with public sector7469requirements. Our AI for regulated customers is in most cases exactly7470the same as what we offer our commercial customers. Our commercial7471Azure cloud is also FedRAMP High authorized and hosts an array of7472State, Local, Federal agencies who utilize this as their primary cloud7473offering today.74747475    Question 8. What protocols do you have in place to work with7476government agencies to rectify any harms or errors when they occur?7477    Answer. Microsoft offers a layered framework that combines7478technical, procedural, and ethical safeguards. We recognize the7479heightened responsibility that comes with deploying AI in the public7480sector, especially where consequential decisions can impact access to7481rights, services, and benefits. Our approach to Responsible AI is7482grounded in principles such as accountability (including human7483oversight and control), transparency, fairness, and reliability &7484safety.7485    For example, we utilized the following techniques:74867487   Human-in-the-Loop Oversight: We ensure that critical7488        decisions involving eligibility or access to services include a7489        human review and override mechanisms.74907491   Incident Response Framework: We work closely with agencies7492        to establish clear escalation paths and remediation protocols7493        for identifying, reporting, and correcting AI-related errors or7494        harms.74957496   Bias and Fairness Audits: Our models undergo rigorous pre-7497        deployment and ongoing audits to detect and mitigate bias,7498        especially in sensitive use cases.74997500   Transparent Documentation: We provide model cards, data7501        sheets, and decision traceability to support explainability and7502        accountability.75037504    Microsoft also provides access to our responsible AI dashboard,7505which enables agencies to monitor fairness, accuracy, and error rates7506across demographic groups. We are committed to continuous collaboration7507with government partners to ensure AI systems are safe, equitable, and7508aligned with public values. We welcome the opportunity to co-develop7509governance frameworks tailored to your mission needs.7510Business Partnerships7511    Question 9. When you make your AI systems available to other users/7512deployers, what are the types of issues you agree on?7513    Answer. Microsoft works closely with its customers, the deployers7514of its AI systems, to ensure that AI technologies are used responsibly,7515safely, and in line with legal requirements. When Microsoft makes an AI7516system available to a customer, both parties agree on key issues of7517responsible AI use, Microsoft provides extensive information and7518guidance to the customer, and Microsoft takes proactive steps to7519prevent misuse of the AI services. These are codified in Microsoft's7520Terms of Use, Acceptable Use Policy, and the Microsoft AI Services Code7521of Conduct, to which customers must adhere. In addition, some services,7522which have a higher risk of misuse, are available only through a7523Limited Access Program.75247525    Question 10. What information do you provide to those other7526parties/deployers?7527    Answer. Microsoft provides extensive information, tools, and7528guidance to customers deploying its AI systems, so they can understand7529how to use the technology responsibly and effectively. This includes7530detailed documentation and transparency notes that explain how AI7531technology works, its capabilities and limitations, and how to achieve7532the best results. We also provide usage guidelines, best practices, and7533responsible AI resources to guide deployers in safe implementation7534(e.g., recommendations for human oversight, testing, and fairness7535checks).7536    Additionally, we build in safety features like content filtering7537and abuse detection models directly into our AI services and offer7538ongoing support, including technical support channels and updates, to7539help customers deploy AI correctly. This helps ensure deployers7540understand the system and have the tools to uphold the agreed7541principles.75427543    Question 11. What step does your company take any steps to reduce7544the likelihood that, for instance, those downstream users don't use7545your services in ways that could harm people or that violate your terms7546of use?7547    Answer. We take active steps to reduce the risk of downstream7548misuse of our AI services. Our services have built-in safety controls7549at multiple levels (the model, the API service, and the application) to7550automatically filter or block harmful content. We enforce our terms of7551use, including by pursuing legal action against actors who try to7552bypass safety measures, such as cybercriminals who intentionally7553develop tools specifically designed to circumvent the safety guardrails7554to create offensive and harmful content. Additionally, over a year ago,7555we implemented security policies to include, if we observe a nation-7556state actor, cybercriminal or other malicious actor using our AI tools7557or services, we will disrupt and disable them immediately, notify any7558service providers they may be using and share our learnings with the7559public and stakeholders to improve the AI ecosystem. Staying ahead of7560threat actors in the age of AI | Microsoft Security Blog Furthermore,7561we continually strengthen our guardrails by learning from new threats7562and updating our safety systems, and we work with partners and provide7563channels for users to report abuses. These measures collectively help7564ensure that even after deployment, our AI services are used in ways7565that do not harm people or violate the agreed-upon terms.7566AI Hallucinations7567    Recent reporting suggests that generative AI hallucinations are7568getting worse as the technologies become more powerful. Hallucinations7569can lead to great harm in certain scenarios, such as when assessing job7570applications, or even more dangerously in the national security7571context.75727573    Question 12. Do you agree that we need guardrails to ensure that AI7574tools are not used or misused in ways that could cause harm to people?7575    Answer. Yes. We agree that thoughtful, risk-based guardrails are7576necessary for advancing safe, trustworthy, and responsible AI.75777578    Question 13. What steps is your company taking to address this7579issue?7580    Answer. Over several years, Microsoft has developed a structured7581approach to responsibly releasing generative AI applications, guided by7582a ``map, measure, and manage'' framework. At each stage of this7583process, we've embedded best practices, guidelines, and tools informed7584by real-world experience. A key focus has been addressing the risk of7585hallucinations, also known as ungroundedness, where AI models generate7586plausible but unsupported content. As part of this comprehensive7587approach, product teams are equipped with centralized tools to evaluate7588the likelihood of ungrounded outputs and are supported with design7589patterns and mitigation strategies tailored to their specific7590applications.7591    A key example of how these risk mitigation practices work is the75922023 release of Copilot Studio, which harnesses generative AI to enable7593customers without programming or AI skills to build their own copilots.7594One of the key risks for this product is groundedness, and, as with all7595generative applications, the Copilot Studio engineering team mapped,7596measured, and managed risks according to our governance framework prior7597to deployment. By improving groundedness mitigations through metaprompt7598adjustments, the Copilot Studio team significantly enhanced in-domain7599query responses, increasing the in-domain pass rate from 88.6 percent7600to 95.7 percent. This means that when a user submits a question that is7601in-domain--or topically appropriate--copilots built with Copilot Studio7602are able to respond more accurately. This change also resulted in a7603notable 6 percent increase in answer rate within just one week of7604implementation. In other words, the improved groundedness filtering7605also reduced the number of queries that copilots declined to respond7606to, improving the overall user experience.7607    The team also introduced citations for outputs, so copilot users7608have more information about the source of information included in AI-7609generated outputs. By amending the safety system message and utilizing7610content filters, the Copilot Studio team improved citation accuracy7611from 85 percent to 90 percent. Following the map, measure, and manage7612framework and supported by robust governance processes, the Copilot7613Studio team launched an experience where customers can build safer and7614more trustworthy copilots.7615    Just as we measure and manage AI risks across the platform and7616application layers of our generative products, we empower our customers7617to do the same. We offer features to our customers that detect7618ungrounded statements within generative AI outputs in applications7619using grounded documents, such as Q&A Copilots and document7620summarization applications. Groundedness detection finds ungrounded7621statements in AI-generated outputs and allows the customer to implement7622mitigations such as triggering rewrites of ungrounded statements.7623                                 ______76247625   Response to Written Questions Submitted by Hon. Tammy Baldwin to7626                               Brad Smith7627    Question 1. I serve as Ranking Member of the LHHS subcommittee of7628the Appropriations Committee where we are tasked with ensuring we are7629investing in the education of the next generation of American workers.7630Part of Microsoft's investment in the Mount Pleasant data center7631includes partnering with Gateway Technical College to build a Data7632Center Academy. Can you share why Microsoft has decided to invest in7633local STEM education?7634    Answer. Microsoft is committed to local workforce development. This7635initiative is designed to train over 1,000 students in 5 years,7636equipping them with the skills needed for careers in IT and data center7637operations. By aligning infrastructure investment with education,7638Microsoft aims to ensure that local communities benefit directly from7639the economic opportunities created. It also supports the company's7640broader goal of building a diverse, future-ready workforce in regions7641poised for tech growth.76427643    Question 2. At the end of 2023, Microsoft entered into a first of7644its kind partnership with AFL-CIO to ensure workers are at the table in7645the development and implementation of artificial intelligence. I7646believe it is critically important that technology advances in a way7647that enhances our workforce instead of eliminating jobs. Can you share7648why Microsoft decided to pursue this partnership and what you have7649gained from it?7650    Answer. Microsoft partnered with the AFL-CIO to ensure that workers7651have a voice in how AI is developed and deployed. The partnership7652focuses on expanding access to AI education, incorporating labor7653feedback into product design, and shaping public policy that supports7654inclusive economic opportunity. It reflects Microsoft's broader7655commitment to responsible AI and to building a future of work that7656benefits everyone. Microsoft has learned the value of embedding worker7657voice directly into the deployment of AI technologies and reinforced7658the need for inclusive AI skilling strategies.7659                                 ______76607661    Response to Written Questions Submitted by Hon. Jacky Rosen to7662                               Brad Smith7663Adversarial AI7664    The reason cybersecurity experts were able to identify gaps in7665security in DeepSeek's platform was because it is an open-source model.76667667    Question 1. How can Congress protect Americans from future models7668developed by entities affiliated with foreign adversaries that may put7669users' data at risk, whether they're open source or not?7670    Answer. *See the answer below76717672    Question 2. Are there ways users--whether Federal or commercial--7673could safely use AI models developed by companies in adversarial7674nations?7675    Answer. There is a clear demand for this technology. As more and7676more businesses and individuals look to innovate and explore new7677markets, they'll want to consider a wide range of AI model types that7678best fit their needs. Access to the latest AI technologies enables7679innovation, but they should be used in a trusted and safe environment.7680When handling sensitive data or used in high-risk use cases, additional7681measures, such as red teaming, automated assessments, and in-depth7682security reviews, can be used to discover and mitigate potential risks7683in models and the systems they're used in. Consistent with shared7684responsibility practices, we encourage customers to carefully review7685model and system documentation and transparency notes and adopt7686platform safeguards, such as content safety filters, and to conduct7687their own security and safety evaluations tailored for their specific7688use case.7689Model Security7690    Question 3. It's essential we ensure the AI models we use do not7691become another cybersecurity vulnerability. Would voluntary7692cybersecurity standards for large AI models or high-risk models and the7693infrastructure they were trained on be helpful in establishing trust?7694    Answer. Voluntary risk-based cybersecurity standards for AI models,7695particularly cutting-edge models and the infrastructure they're trained7696on--would be a valuable step toward building trust.7697    Previous waves of technology have highlighted that to drive7698innovation and adoption of new technologies, users must both trust in7699how technology itself performs and trust that the technology can be7700used successfully, safely and securely. Cybersecurity frameworks and7701standards provide a foundation for that trust. For the past decade,7702risk-based cybersecurity standards developed by organizations such as7703NIST and ISO/IEC have helped public and private sector stakeholders7704manage cybersecurity risks effectively, regardless of their size or7705maturity. These standards also support compliance and assurance schemes7706that deliver both economic and security benefits.7707    In our public comments to the White House Office of Science and7708Technology Policy (OSTP) in its Request for Information (RFI) for the7709Development of an AI Action Plan, Microsoft encouraged the Federal7710government to avoid developing duplicative policies by leveraging7711existing risk-based and outcome-focused cybersecurity standards, such7712as the NIST Cybersecurity Framework and the Secure Software Development7713Framework (SSDF). These standards offer the flexibility needed to7714streamline regulation, promote consistency, and foster innovation and7715cross-border collaboration.7716    Unlike prescriptive cybersecurity requirements, risk-based7717standards offer flexibility and agility to streamline regulations,7718drive consistency, and incentivize innovation and cross-border7719commerce. Developing new standards from scratch is a resource-intensive7720process, and the Federal government can save substantial resources7721using and building existing risk-based standards while still ensuring7722robust cybersecurity measures are in place. In our comments, we also7723encourage the administration to invest in NIST, as they play a critical7724role in informing the global AI conversation, bringing U.S.7725perspectives to international standardization bodies like ISO/IEC.7726Alignment and interoperability between U.S. cybersecurity and7727international standardization approaches is important for the growth of7728cross-border commerce and trust in American AI.7729AI Skills7730    Question 4. You mentioned the importance of digital skills in your7731testimony. Can you discuss how it might hurt the U.S.'s ability to7732compete with China if we don't leverage congressionally-mandated7733Federal programs like those created under the Digital Equity Act, which7734were explicitly designed to help Americans build digital skills, like7735teaching seniors, small businesses, and veterans how to use AI?7736    Answer. Americans of all ages and backgrounds will need AI skills7737to compete in this new world of work. A key opportunity for most people7738will be to develop an AI fluency that will enable them to use AI in7739their jobs, much like they use laptops, smartphones, software7740applications, and the Internet today. China is home to some of the7741world's most talented computer science researchers. They also lead the7742world in graduating STEM students. Therefore, it is critical the United7743States, government, private sector, and non-profits, to invest in AI7744and digital skilling. For example, we launched the AI for Community7745Colleges program, in collaboration with the American Association of7746Community Colleges, designed to empower both students and educators by7747providing valuable resources and support. The program offers AI7748training for faculty and staff tailored to all skill levels, ensuring a7749comprehensive understanding of AI concepts. It will deliver AI-focused7750curriculum that equips students with in-demand skills to meet regional7751workforce needs, as well as AI technology to enhance various7752departmental functions at community colleges. By helping educators7753integrate AI skilling directly into their classrooms, the program7754ensures students are well-prepared to enter the workforce as AI-ready7755professionals, addressing the growing demand for AI expertise across7756industries.7757AI For Spectrum7758    Artificial Intelligence has the potential to dramatically improve7759how we manage and utilize spectrum, particularly unlicensed bands like7760those used for Wi-Fi.77617762    Question 5. What are some of the most promising ways AI can be7763applied to enhance dynamic spectrum access, reduce interference, and7764optimize performance in congested environments--and what should7765Congress do to support those efforts?7766    Answer. There are companies that are applying machine learning and7767AI to create automated tools to better manage spectrum utilization in7768various frequency bands through shared use. There are also universities7769examining some of the more fundamental questions associated with7770dynamic spectrum sharing. Presumably, some of this research is7771exploring how AI tools can be applied. However, Microsoft is not7772involved in these efforts.77737774    Question 6. The Citizens Broadband Radio Service (CBRS) model has7775demonstrated how dynamic sharing between Federal and non-federal users7776can unlock valuable spectrum for innovation. How can AI enhance and7777expand these types of spectrum sharing frameworks by enabling more7778agile, real-time spectrum coordination, and what steps should the7779Federal government take to accelerate the development of AI-powered7780spectrum management tools?7781    Answer. Although Microsoft participated in multiple CBRS7782proceedings that go back over a decade, in recent years the company has7783not been engaged.7784Need for Broadband and Wi-Fi7785    Question 7. Wi-Fi is the foundation of connectivity in our homes,7786schools, hospitals, and workplaces. As AI becomes more embedded in7787applications across sectors--from diagnostics and patient monitoring to7788smart factories and personalized education--how critical is it that we7789continue to invest in broadband infrastructure across the U.S. and7790robust, high-capacity Wi-Fi networks to realize AI's full economic and7791social potential?7792    Answer. Microsoft agrees that Wi-Fi-enabled devices play an7793essential role in today's communications networks and will continue to7794do so in the future. The Federal Communication Commission's (FCC)7795decisions to authorize unlicensed Low Power Indoor (LPI) and Very Low7796Power (VLP) devices across the entire 5925-7125 Megahertz (6 GHz) band7797and higher-power Standard Power (SP) devices under control of an7798Automated Frequency Coordination (AFC) system over portions of the 67799GHz band, allow for multiple, large bandwidth channels that serve as7800large on-ramps for individuals and enterprise users to access broadband7801services, and which is a prerequisite for high-capacity low-latency Wi-7802Fi networks. FCC authorization of LPI, VLP, and SP (under control of an7803AFC) devices has set the stage for significant private sector7804investments. With respect to AI enabled devices, several of Microsoft's7805more recently released Copilot + PC computer models incorporate Wi-Fi 77806radios that feature 320 Megahertz ultra-wide channels. Along with our7807customers, we are learning in real-time how AI embedded systems and7808services can best leverage high-capacity low latency Wi-Fi connections.7809AI T&E Workforce7810    A key factor in ensuring the U.S. continues to lead the world in7811the AI race is by ensuring the AI we develop is the best and therefore7812the most trustworthy. Validating model outputs is an important step in7813establishing trust. Right now, however, the U.S. has neither the7814standards nor the trained workforce to evaluate AI models to establish7815that we can trust model outputs.78167817    Question 8. What should Congress consider to incentivize and grow7818the AI test and evaluation workforce?7819    Answer. Congress should consider creating AI upskilling sector-7820based collaboratives, where governments fund efforts that bring7821together companies that use the same or similar AI technologies to help7822train workers more efficiently and at scale, which would be especially7823beneficial to subject matter experts that may not otherwise have the7824infrastructure to establish AI training programs. Additionally, we7825recommend advocating for expanding, streamlining and promoting Section7826127, which allows employers to provide tax-free educational assistance7827to employees pursuing education while working. As part of any effort to7828expand or streamline Section 127, legislation could also highlight or7829further incentivize employers using this benefit for AI upskilling.7830There may be an opportunity to further promote Section 127 benefits,7831which are often underutilized, by highlighting the ways in which it can7832address AI skilling needs.78337834    Question 9. How can Congress support more interdisciplinary7835approaches to testing and evaluating AI? For example, how do we ensure7836a model being used in a healthcare setting has been evaluated both by7837experts in the model technology, but also experts in the healthcare7838setting in which it will be deployed?7839    Answer. Microsoft has consistently emphasized that cross-7840disciplinary testing of AI--specifically through testing specific to a7841deployment setting--is essential to secure and trustworthy AI use.7842During our own product development, we conduct stress testing, or red7843teaming if necessary, at both the model and the application layer. Red7844teaming the model helps to identify how a model can be misused, scope7845its capabilities, and understand its limitations. These insights not7846only guide the development of platform-level evaluations and7847mitigations for use of the model in applications but can also be used7848to inform future versions of the model. Application-level AI red7849teaming takes a system view, of which the base model is one part. This7850helps to identify failures beyond just the model, by including the7851application specific mitigations and safety system. Red teaming7852throughout AI product development, when appropriate, can surface7853previously unknown risks, confirm whether potential risks materialize7854in an application, and inform measurement and risk management. The7855practice also helps clarify the scope of an AI application's7856capabilities and limitations, identify potential for misuse, and7857surface areas to investigate further.7858    Congress can assist these efforts by supporting and funding NIST,7859which is currently considering scoping for workstreams to develop7860methods and metrics for AI testing, evaluation, verification, and7861validation.7862                                 ______78637864   Response to Written Question Submitted by Hon. John Fetterman to7865                               Brad Smith7866    Mr. Smith, I'm a big supporter of renewable energy, and that7867includes nuclear. Whenever we talk about the energy transition, as we7868discussed at this hearing, my focus has been clear: making sure7869ratepayers in Pennsylvania aren't hurt. The Washington Post has7870reported that increasing electricity demand from data centers is7871jacking up residential power bills by 20 percent.\13\ That's7872unacceptable. These data centers don't even offer long lasting, stable7873jobs: the new jobs are in the construction phase, but the higher7874utility rates last forever. I've been tracking the plan to reopen Three7875Mile Island to power Microsoft's data center energy needs. I appreciate7876innovative plans, but Pennsylvanians come first.7877---------------------------------------------------------------------------7878    \13\ https://www.washingtonpost.com/business/2024/11/01/ai-data-7879centers-electricity-bills-google-amazon/78807881    Question. Will you commit to me--as you did at the hearing--that7882your power purchase agreement with Constellation will not raise7883electricity rates for PA households?7884    Answer. The power purchase agreement that Microsoft entered into7885with Constellation Energy will add 835 MW of electricity to the PJM7886region. By entering into this power purchase agreement, Microsoft is7887guaranteeing a customer for this power produced by the nuclear unit at7888the Crane Clean Energy Center and fully responsible for the costs of7889the energy and capacity from this facility.7890                                 ______78917892Response to Written Questions Submitted by Hon. Lisa Blunt Rochester to78937894                               Brad Smith7895Cybersecurity and AI7896    Mr. Smith, I know that many tech companies like yours see AI agents7897as a big part of AI advancement.7898    But AI agents also can contain sensitive data about its users, like7899a flight agent containing payment information and location data. This7900data needs to be secure from cyberattacks.7901    Also, there is potential for AI agents to be used by cybercriminals7902to orchestrate attacks more quickly and at a far larger scale than7903humans could.79047905    Question 1. I know that Microsoft has a good amount of visibility7906into global cybersecurity threats. How would you assess the current7907state of AI cybersecurity, and what concrete steps should this7908committee consider to strengthen the cyber-protection of AI agents?7909    Answer. Earlier this month at Microsoft Build, our annual developer7910conference, we outlined our vision for a world where AI agents make7911decisions and perform tasks across users, teams, or organizations. To7912realize our vision, AI agents must be both capable and secure. Today's7913threat landscape--from the unprecedented volume of ransomware attacks7914by cybercriminals to the sustained cyberespionage and attacks from7915state-sponsored adversaries--already demands a strong, coordinated7916response. AI agents, especially those interacting with sensitive data,7917introduce new risks that must be addressed proactively and through7918trusted multistakeholder partnerships.7919    AI agents can operate autonomously, make real-time decisions,7920interact with external tools and data, and even collaborate with other7921agents. This has the potential to transform industries, from optimizing7922energy grids to coordinating fleets of autonomous vehicles. More than7923230,000 organizations--including 90 percent of the Fortune 500--have7924already used Copilot Studio to build AI agents and automations.7925    These powerful capabilities also introduce new risks. AI agents can7926be manipulated through prompts or data sources to perform harmful7927actions--like profiling employees, crafting targeted phishing e-mails,7928or leaking sensitive information. In multi-agent systems, a single7929compromised agent can trigger cascading failures. For example, a hacked7930warehouse robot could disrupt an entire supply chain. An attacker could7931hide secret instructions in a document used for training of or accessed7932by a public-facing agent. When a secure AI agent later interacts with7933the public-facing agent, the embedded instructions can trick it into7934bypassing its safeguards and leak sensitive data. These risks are7935amplified because these systems often operate without direct human7936oversight, raising important questions about accountability.7937    We would welcome the opportunity to engage with you further on this7938topic, and hope to see the Committee's support for the following:79397940   Emphasize robust cybersecurity foundations, such as secure-7941        by-design-and default practices and Zero Trust Architecture for7942        all AI systems--including agents.79437944   Support voluntary, risk-based AI cybersecurity standards7945        that are adaptable, technically grounded, and internationally7946        aligned.79477948   Remove barriers to cloud adoption, recognizing the security,7949        scalability and efficiency benefits cloud services provide for7950        AI agent deployment.79517952   Encourage open, secure protocols like Agent2Agent (A2A) and7953        Model Context Protocol (MCP) to enable safe and interoperable7954        agent-to-agent and agent-to-tool collaboration.79557956   Invest in public-private collaboration to share threat7957        intelligence, simulate adversarial scenarios, and evolve best7958        practices.7959AI and Competition7960    Mr. Smith, AI is becoming integrated into our critical economic and7961societal infrastructure, with McKinsey stating that long-term AI7962opportunity could be about $4.4 trillion in added productivity growth7963potential from corporate use cases.7964    But vendor lock in could be a real issue, where an AI vendor7965dramatically falls behind the competition and leaves its client with a7966vastly inferior product, which could threaten key industries the AI7967product operates in.79687969    Question 2. Do you have any plans or strategies regarding7970mitigating lock-in for your AI products operating in critical sectors,7971like the financial and medical sectors, to prevent potential lock-in7972effects that might harm these critical sectors and the folks therein?7973    Answer. Today's AI ecosystem is open, evolving, and increasingly7974decentralized. Microsoft is working to ensure it stays that way.7975    We are seeing a surge of innovation across sectors, with new7976models, tools, and platforms emerging regularly. This is especially7977true in financial services and public health, where AI's potential is7978still be discovered and applied. From fraud detection to disease7979surveillance, the applications are expanding rapidly. And so are the7980choices available to customers.7981    At Microsoft, we are committed to supporting that diversity. Azure7982hosts a wide range of AI models, including those from OpenAI, Meta,7983Mistral, and open-source communities; we recently added xAI's Grok 37984and Grok 3 Mini into the Azure AI Foundry. Customers can also bring7985their own models, fine-tune them, or use pre-trained ones. They can run7986models in Azure, on-premises, or across multiple clouds. We support7987that flexibility because we believe it's essential to trust and7988innovation.7989    We also recognize that many of our customers operate in regulated7990environments. That's why we invest in interoperability, portability,7991and compliance--so they can move workloads as needed, with minimal7992friction and constraint.79937994                               [all]