Editor’s notice: On Thursday, Could 8, Microsoft Vice Chair and President Brad Smith testified earlier than the Senate Commerce Committee. To view the proceedings, go to the committee’s web site.
Profitable the AI Race:
Strengthening U.S. Capabilities in Computing and Innovation
Written Testimony of Brad Smith
Vice Chair and President, Microsoft Company
Senate Commerce Committee
Chairman Cruz, Rating Member Cantwell, and Members of the Committee,
Thanks for the chance to testify on the crucial problem of synthetic intelligence. I’m Brad Smith, the Vice Chair and President of Microsoft Company.
AI has the potential to develop into essentially the most great tool for individuals ever invented. Like the overall function applied sciences that preceded it, resembling electrical energy, machine instruments, and digital computing, AI will impression each a part of our economic system. It can form not simply how we work and stay, however how we compete, prosper, and keep safe as a nation between now and the center of this century.
The discover for this listening to aptly refers to an “AI race.” I want to speak at the moment about what is required to win this race.
The AI race includes each expertise and economics. It requires each innovation and diffusion. It’s each a dash and a marathon. The nation can win a lap however lose the race if it fails to deliver collectively all of the elements wanted for fulfillment.
It’s a race that no firm or nation can win by itself.
To win the AI race, the US might want to assist the non-public sector at each layer of the AI tech stack. The nation might want to associate with American allies and mates around the globe.
In my testimony at the moment, I’ll concentrate on three strategic priorities the place this Congress and the federal authorities will make a distinction.
First, the nation should win the AI innovation race. It will require huge datacenters and AI infrastructure that want federal assist to increase and modernize {the electrical} grid on which they rely. The nation should recruit and practice expert labor like electricians and pipefitters which can be in brief provide. All of us should summon the perfect of our researchers at nationwide labs and universities, supported by federal primary analysis applications and partnerships which have develop into the envy of the world. We might want to proceed to excel in transferring progressive concepts from educational labs into firms and new merchandise. And we might want to assist AI builders with open and broad entry to public knowledge.
Second, the nation should win the AI diffusion race. It will require that we promote broad AI adoption that may allow productiveness development throughout each sector of the economic system. Greater than something, this requires new initiatives to advertise the AI skilling of the American workforce. It will contain primary AI fluency in our faculties and new AI coaching applications in our group schools. It can additionally embrace superior AI training that may signify the subsequent technology of laptop science levels, organizational expertise that will likely be mastered within the nation’s enterprise faculties, and new programs within the nation’s regulation faculties. When mixed, these will allow firms, non-profits, and authorities businesses alike to place AI to efficient use. Governments on the federal, state, and native ranges can then assist speed up this diffusion by adopting AI companies to enhance the effectiveness and effectivity of the companies they supply to the general public.
Third, the US should export AI to American allies and mates. No firm or nation is so highly effective that it may well grasp the way forward for AI with out mates. The US and China are competing not solely to innovate however to unfold their respective applied sciences to different nations. This a part of the race probably will likely be gained by the quickest first mover. The US wants a wise export management technique that protects our nationwide safety whereas assuring different nations that they are going to have dependable and sustained entry to crucial American AI parts and companies. Maybe as a lot as something, this requires that we collectively maintain worldwide belief in our merchandise, our firms, and the nation itself.
AI as a Common Function Know-how
Economists generally put applied sciences into two classes, common function applied sciences and single-purpose instruments. Most issues on the earth are single-purpose instruments, like a smoke detector or a garden mower. They do one factor very properly. However over the course of historical past, sure so-called common function applied sciences impression and generally even redefine nearly each sector of the economic system. Electrical energy is the prototypical instance, as a result of when you concentrate on it, electrical energy modified the way in which each financial sector works.
The important thing to mastering the way forward for AI begins partly by understanding the function expertise has performed prior to now. The previous three centuries have introduced the world three industrial revolutions, every pushed by these common function applied sciences. First, it was iron working in the UK, beginning within the 1700s. After which it was electrical energy and machine instruments within the 1800s, when the US overtook the UK by placing these applied sciences to work extra broadly than every other nation. After which there was the third industrial revolution over the past 50 years, pushed by laptop chips and software program.
With out query, being a worldwide chief in advancing a common function expertise offers a rustic a serious edge. However one lesson of historical past is that the nations that profit essentially the most and advance the quickest will not be essentially the nations the place the expertise is invented. Somewhat, it’s the place the expertise is subtle – or adopted – essentially the most rapidly and broadly. That is for good cause. If a expertise improves productiveness and modifications each a part of an economic system, then the nation that makes use of it essentially the most broadly and rapidly will profit essentially the most.
This each frames and defines the AI alternative and problem for the US. As a nation, we have to focus each on advancing innovation and driving diffusion, each domestically and as a number one American export.
The AI Tech Stack
The important thing to driving each innovation and diffusion is to acknowledge that AI, like all common function applied sciences, is constructed on what we within the business name a tech stack – a stack of applied sciences which can be used collectively. That is true for each nice common function expertise. You may see this, for instance, if we return in time and take into consideration electrical energy. Thomas Edison first succeeded in 1878 in utilizing electrical energy to mild a lightbulb. However the illumination of lights throughout a metropolis rapidly required the development of energy vegetation, the gas to run them, the creation of {an electrical} grid, the standardization of circuits, and a variety {of electrical} home equipment past the lightbulb itself. Briefly, a tech stack for electrical energy.
Synthetic intelligence equally is constructed on an AI tech stack. Basically, it’s divided into three layers, infrastructure, the platform layer, and functions. You may see this illustrated beneath.
The infrastructure layer is huge. Microsoft is spending greater than $80 billion this fiscal 12 months on the capital funding wanted for this layer, with greater than half this quantity being spent in the US. This goes to purchasing land, investing in electrical energy and broadband connectivity, procuring chips like GPUs, and putting in liquid cooling. These result in the development of datacenters – or typically datacenter campuses with many buildings with doubtlessly lots of of 1000’s of computer systems. This infrastructure helps each the coaching of recent AI fashions and their deployment, to allow them to be used for AI-based companies around the globe.
On high of this infrastructure, there’s the platform layer. The center of this layer consists of AI basis fashions, together with frontier fashions created by firms like OpenAI, in addition to open supply and different fashions from all kinds of different companies – together with Anthropic, Google, Mistral, DeepSeek, and Microsoft itself. The platform layer depends on knowledge to coach and floor fashions. And it features a new technology of software-based AI platform companies which can be used to assist construct AI functions.
In the end, each the infrastructure and platform layers assist the functions layer. These are units and software program functions that use AI to ship higher companies to individuals. ChatGPT and Microsoft’s Copilot are each examples of AI functions. One of many wonderful issues concerning the functions layer is it’s not simply firms – giant or small or established or startup – which can be creating AI functions. It’s all people. It’s researchers utilizing new AI-infused functions to alter drug discovery. It’s non-profits altering the way in which they ship companies. It’s academics utilizing AI as a software to enhance the way in which they put together materials for a classroom. It’s governments making all the pieces from the submitting of a tax return to the renewal of a driver’s license simpler and extra environment friendly.
To construct a brand new AI economic system, it’s crucial to get all three of those layers working and to get a flywheel turning throughout the ecosystem. It’s important to construct the infrastructure layer so individuals can develop and deploy the fashions on the platform layer. It’s important to make use of the AI fashions so that individuals will construct the functions on high of them. And it’s important for patrons to undertake the functions, so the market can develop, and drive elevated funding to increase the infrastructure additional. The method repeats itself. That is how a brand new economic system is born.
Success Requires an Whole Ecosystem
The flywheel impact makes clear that success requires not solely nationwide progress at one layer of the tech stack, however at each layer. That’s what the non-public sector at the moment is pursuing in the US higher than in every other nation. And it’s what this Congress and the Government Department may help assist with a method that promotes each AI innovation and diffusion up and down this stack.
Nationwide AI management requires not solely success by a number of firms, however by many. In the present day’s panel, involving main companies resembling OpenAI, AMD, CoreWeave, and Microsoft, displays necessary slices of the brand new AI economic system. The AI economic system requires a multifaceted and built-in ecosystem that features “Huge Tech” and “Little Tech,” startups and extra established companies, open supply and proprietary builders, suppliers and clients, companies that create knowledge and companies that eat it, all working collectively. Governments as each regulators and main AI adopters have crucial roles to play.
Commentators generally concentrate on the tensions between completely different contributors on this tech ecosystem. These deserve consideration. What’s typically missed is that the completely different contributors additionally depend upon one another. And because of this the completely different contributors to the AI ecosystem all have to be wholesome.
A big expertise firm like Microsoft has a singular alternative – and duty – to associate with and assist the contributors at each stage of the tech stack. We attempt to advance not simply innovation however an financial structure, enterprise fashions, and accountable practices that may assist develop the AI market on a long-term foundation. Not only for the US, however the nation’s mates and allies.
Profitable the Innovation Race
Though the AI economic system is being constructed largely by the non-public sector, authorities insurance policies and initiatives must play a crucial function. This begins with work wanted to assist gas innovation. A number of areas deserve explicit consideration on this listening to.
Energy the expansion of datacenters
Simply as you may’t have dependable electrical energy in your house and not using a powerplant, you may’t have AI with out datacenters and AI infrastructure. And these datacenters require an unlimited provide chain to assemble and enormous quantities of electrical energy to function.
America’s superior economic system depends on 50-year-old infrastructure that can’t meet the growing electrical energy calls for pushed by AI, reshoring of producing, and elevated electrification. The US might want to put money into extra transmission and vitality sources, onshore our provide chains, and modernize our electrical grid to assist forecasted will increase in electrical masses. Microsoft is investing in these areas itself.
We urge the federal authorities to streamline the federal allowing course of to speed up development in all these areas. The present federal allowing processes typically contain a number of businesses and sophisticated, unpredictable, multi-year evaluations. This hinders progress. The federal authorities ought to take quick steps to determine dependable, cheap, and clear timelines for allowing selections. This can be performed by standardizing federal allowing processes and designating a lead company to shepherd the permits via the method. Additional, the allowing businesses ought to make the most of AI and digital instruments to enhance timelines and transparency for candidates and make sure the allowing businesses have fast entry to data to help them of their evaluation and decision-making course of.
We have been happy to see President Trump’s current Government Order, “Updating Allowing Know-how for the 21st Century,” directing businesses to make most use of expertise within the environmental evaluation and allowing course of. The Congress also needs to look to the Federal-State Trendy Grid Deployment Initiative as a confirmed program that may be leveraged to ship outcomes.
That is simply the beginning of what’s wanted to modernize and increase America’s vitality grid. We have to acknowledge that new investments within the grid are simply as necessary at the moment as they have been a century in the past, when the US led the world in non-public and public sector assist for electrical energy.
Develop the AI Infrastructure workforce
Maybe the only greatest problem for knowledge middle growth in the US is a nationwide scarcity of individuals – together with expert electricians and pipefitters. Electricians, for instance, are important to datacenter development, putting in a posh system {of electrical} panels, transformers and backup energy techniques. We’ve employed 1000’s of electricians throughout the nation, together with in Arizona, Georgia, Virginia, Washington, and Wisconsin. However the US doesn’t have sufficient electricians to fill the rising demand. We estimate that over the subsequent decade, the US might want to recruit and practice half 1,000,000 new electricians to fulfill the nation’s rising electrical energy wants. We want a nationwide technique to make sure we meet this chance for American employees.
These are good jobs that may present nice long-term careers for individuals throughout the nation. We advocate making current federal training and coaching funds, in addition to tax incentives, accessible to scale up these alternatives. These may embrace concentrating on present federal apprenticeship investments in areas which have recognized main AI infrastructure initiatives and supporting current coaching facilities to rapidly improve the variety of registered apprenticeships targeted on electricians.
We commend President Trump’s current Government Order, “Making ready People for Excessive-Paying Expert Commerce Jobs of the Future,” for highlighting the significance of expert trades within the constructing of AI infrastructure and for paving the way in which to fulfill this second. As federal businesses work to implement the order, it will likely be crucial that business forecasters and union coaching facilities work collectively to maximise impression.
In the end, we want new steps at each stage of presidency and in communities throughout the nation. For instance, we have to do extra as a nation to revitalize the economic arts and store lessons in American excessive faculties. This needs to be a precedence for native college boards and state governments. Equally, the nation’s group schools might want to do extra to assist a nationwide initiative to assist practice a brand new technology of expert labor, together with electricians and pipefitters.
Spend money on AI analysis and improvement
To uphold America’s place as a worldwide scientific chief, it’s crucial to boost federal funding in elementary scientific analysis. The US boasts a storied historical past of using public-private partnerships. The choices made many years in the past to publicly fund analysis infrastructure and supply monetary assist to gifted scientists and entrepreneurs paved a pathway to American technological management. By way of federal, state and native authorities initiatives, investments have been made in regional economies and applications, betting on the ingenuity of the American individuals. Notable incubators of the 20th century – resembling Bell Labs and the community of federal nationwide laboratories – have been the results of deliberate efforts to unite business, authorities, and academia to propel scientific development. We should deploy an analogous technique at the moment for AI and quantum applied sciences. Investments in these areas are crucial to advancing the event of progressive technological options that tackle complicated world challenges.
To outcompete nations like China, which have considerably boosted their analysis and improvement (R&D) investments, the US should speed up strategic investments in scientific analysis for future applied sciences. Specialists predict China will proceed to speculate substantial sources in next-generation applied sciences resembling AI, superior manufacturing, clear vitality, quantum computing, and semiconductors over the subsequent decade.
Because the Second World Warfare, America’s technological innovation has been pushed by R&D based mostly on two crucial elements that the remainder of the world has each studied and envied. The primary is sustained assist for primary analysis. Whereas a number of tech firms make investments substantial sums in primary analysis, as we do via Microsoft Analysis (MSR), most world-leading primary analysis is pursued by teachers at American universities, typically based mostly on funding from the Nationwide Science Basis and different federal businesses. Pushed by curiosity quite than a revenue motive, this analysis typically results in sudden however profound discoveries which can be printed publicly.
The second ingredient is a sustained dedication to investments in product improvement by firms of all sizes. The US, greater than every other nation, has mastered the method of transferring new concepts rapidly from universities to the non-public sector. This success rests on wholesome investments in each R and D, recognizing that primary analysis is usually publicly funded and sometimes in universities, whereas product improvement is robustly and privately funded via firms. It’s the mix of the 2 that makes American R&D so profitable.
In 2019, President Trump authorised an government order designed to strengthen America’s lead in synthetic intelligence. It rightly targeted on federal investments in AI analysis and making federal knowledge and computing sources extra accessible. Six years later, the President and Congress ought to increase on these efforts to assist advancing America’s AI management. Extra funding for primary analysis on the Nationwide Science Basis and thru our universities is one good place to begin.
Guarantee public knowledge is open and accessible
Information is the gas that powers synthetic intelligence. The standard, amount, and accessibility of information straight determines the power and class of AI fashions. Whereas the web has been a serious supply of coaching knowledge, the federal authorities stays one of many largest untapped sources of high-quality and high-volume knowledge. But at the moment, many of those datasets are both inaccessible or not usable for AI improvement.
By making authorities knowledge available for AI coaching, the US can considerably speed up the development of AI capabilities, driving innovation and discovery. Opening entry to those datasets would enable for the evaluation of themes, patterns, and insights throughout broad datasets, propelling the nation to the forefront of world AI improvement.
Importantly, accessible public knowledge ranges the enjoying subject. It empowers not solely giant firms however startups, educational establishments, and nonprofits to coach and refine AI fashions. This fosters a extra aggressive and inclusive AI ecosystem, the place innovation is pushed by concepts and ingenuity – not simply proprietary knowledge.
Compared, nations like China and the UK are already investing closely of their knowledge sources, recognizing the financial and strategic worth of national-scale knowledge administration. China’s complete system to handle datasets as a strategic useful resource and the UK’s Nationwide Information Library underscore a rising world pattern of treating knowledge as a standard good for financial competitiveness.
Profitable the AI Diffusion Race
Historical past teaches us that the true impression of a general-purpose expertise just isn’t measured solely by the caliber of its main innovations, however by how rapidly, broadly, and successfully these are adopted throughout society. However the actuality is that expertise diffusion takes time, funding, partnerships, and sound public coverage.
The historical past of electrical energy provides an necessary perception for AI. As soon as Thomas Edison proved in 1878 that electrical energy may energy a lightbulb, why would anybody select to take a seat at night time in a room illuminated by a candle or kerosene? But tonight, nearly 150 years later, greater than 700 million individuals on the planet nonetheless stay with out electrical energy of their houses. Diffusion requires not solely nice expertise, however sound economics.
The economics of tech diffusion begin with skilling. International locations must put money into the talents wanted to make use of new expertise, each as people and throughout organizations. It’s simple to underestimate each the function that skilling performs and the necessity for public coverage to assist it. However in every industrial revolution, the nation that finest harnessed the main general-purpose expertise of its time was the nation that expert its inhabitants essentially the most rapidly and broadly.
Talent the American workforce
Within the new AI economic system, People of all backgrounds will want crucial AI expertise to compete. To fulfill the totality of the skilling problem, the nation should pursue a brand new nationwide objective to make AI skilling accessible and helpful for each American. It will require a really broad vary of partnerships and new coverage concepts, spanning throughout geographic, organizational, financial, and political divides.
President Trump’s current government orders targeted on AI training and the workforce present crucial steps in direction of a nationwide skilling technique for AI. The “Advancing Synthetic Intelligence Training for American Youth” EO establishes a transparent coverage to advertise AI literacy by responsibly integrating AI into training for academics and college students. By fostering this early publicity, the nation’s youth will likely be higher positioned for AI-enabled work. Congress can even contemplate leveraging current federal funding to the nation’s college districts to encourage AI studying and literacy in Ok-12 training.
Companies and non-profits have necessary roles to play. At Microsoft, we’re in search of to do our half to fulfill this skilling problem. In 2025 alone, we’re on a path to coach 2.5 million People in primary AI expertise. We’re partnering with the Nationwide Future Farmers of America (FFA) to coach educators in each state to combine AI into the agricultural classroom via our Farm Beats for College students program. We’re partnering with the American Federation of Academics (AFT), the biggest group representing the nation’s educators in America, to ship a co-developed coaching program to 10,000 AFT members. And we’re partnering with the State of New Jersey, Princeton College, and CoreWeave on an AI Hub in New Jersey that may embrace assist for AI training in local people schools.
On the subject of AI skilling, crucial factor we have to do is acknowledge that this can be a crucial subject that’s ripe for consideration, studying, partnership, and innovation. It can have a big impact on broadening entry to this expertise throughout our economic system and society. Generative AI is a brand new and younger expertise. So is our data of the complete extent of want when it comes to AI skilling applications and assist. This can be a first-class precedence that deserves as a lot consideration and assist as innovation in AI expertise itself.
Encourage AI adoption
The federal authorities additionally will play a crucial function in AI diffusion through the use of AI itself. There are alternatives throughout the federal government to make use of AI to enhance the standard and effectivity of public companies for residents.
It’s encouraging to see the current OMB publication of M-Memos targeted on federal authorities use and procurement of AI. Each memos emphasised the significance of eradicating obstacles to innovation, maximizing the usage of domestically developed AI merchandise, and inspiring AI leaders inside the federal authorities to facilitate accountable AI adoption.
We’re seeing exercise within the states as properly. We partnered with the Texas Division of Transportation to launch a six-week pilot program geared toward boosting productiveness and enhancing decision-making throughout varied departments. This system noticed sturdy outcomes with 97 p.c of contributors utilizing the AI digital assistant throughout the pilot, 68 p.c have built-in it into their day by day workflow, and contributors reporting saving a mean of 12 hours every week on routine duties.
Exporting American AI
The flexibility to export our AI is important to sustaining our world competitiveness and guaranteeing that our technological progress advantages not solely our nation, but in addition our allies and companions around the globe. Constructing on current AI diplomacy efforts, the US provides a compelling and trusted worth proposition within the world expertise panorama.
American tech firms, together with Microsoft, are making unprecedented investments in AI infrastructure around the globe. Microsoft alone is constructing AI infrastructure in additional than forty nations, together with areas the place China has targeted its investments. We urgently want a nationwide coverage that gives the fitting steadiness of export controls and commerce assist for these investments.
Whereas the U.S. authorities rightly has targeted on defending delicate AI parts in safe datacenters via export controls, an much more necessary component of AI competitors will contain a race between the US and China to unfold their respective applied sciences to different nations. Given the character of expertise markets and their potential community results, this race between the US and China for worldwide affect probably will likely be gained by the quickest first mover. The US wants a wise worldwide technique to quickly assist American AI around the globe.
This elementary lesson emerges from the previous twenty years of telecommunications gear exports. Initially, American and European firms resembling Lucent, Alcatel, Ericsson, and Nokia constructed progressive merchandise that outlined worldwide requirements. However as Huawei invested in innovation and China’s authorities sponsored gross sales of its merchandise, particularly throughout the growing world, adoption of those Chinese language merchandise outpaced the competitors and have become the spine of quite a few nations’ telecommunications networks. This created the expertise basis for what later turned an necessary problem for the Trump Administration in 2020, because it grappled with the presence of Huawei’s 5G merchandise and their implications for nationwide and cybersecurity.
Early indicators counsel the Authorities of China is focused on replicating its profitable telecommunications technique. China is beginning to supply growing nations sponsored entry to scarce chips, and it’s promising to construct native AI datacenters. The Chinese language correctly acknowledge that if a rustic standardizes on China’s AI platform, it probably will proceed to depend on that platform sooner or later.
Worldwide partnerships will likely be crucial. That is why Microsoft has partnered with entities just like the UAE’s G42 and funding funds like Blackrock and MGX, aiming to lift as much as $100 billion for AI infrastructure and provide chains. American tech firms and personal capital markets are forging stronger ties with key nations and sovereign buyers within the Center East, surpassing earlier efforts to counter Chinese language subsidies in telecommunications and reflecting our dedication to innovation and cooperation. Whereas China’s authorities might subsidize its expertise adoption in growing areas, it would wrestle to match the dimensions and impression of America’s non-public sector investments.
Pragmatic American export management insurance policies are important, balancing safety protections with the power to increase quickly. Defending nationwide safety by stopping adversaries from buying superior AI expertise is essential. Guidelines ought to embrace qualitative requirements for safe datacenter deployments to forestall chip diversion to China and guarantee superior AI companies are safeguarded. We assist the sort of method.
Nonetheless, we’ve got expressed our considerations concerning the quantitative caps imposed on GPU shipments by the interim remaining AI Diffusion Rule issued in January. These place key American allies and companions in a Tier Two class, imposing limits on AI datacenter growth. This consists of nations like Switzerland, Poland, Greece, Singapore, India, Indonesia, Israel, the UAE, and Saudi Arabia. Clients in these nations now concern restricted entry to American AI expertise – doubtlessly benefitting China’s AI sector by turning to options.
The Trump administration has a possibility to revise the rule, eliminating quantitative caps and retaining qualitative requirements. This method ensures American allies and companions stay assured in accessing American AI merchandise.
In the end, we have to acknowledge that nations around the globe will use American AI provided that they’ll belief it. This creates obligations for American firms to develop and deploy AI infrastructure and merchandise in a accountable method that meets native wants. And it requires that nations trust in sustained and uninterrupted entry to crucial AI parts and companies. The US has lengthy constructed a fame for reliable expertise that China has been unable to match. However this fame, like all the pieces that really issues, requires fixed consideration and care.