How the White House Plans to 10x Scientific Productivity | Michael Kratsios | EP #276

1h 12m 76K views 4 Aug 2026

Is it worth your time?

You get a detailed account of specific government programs, export strategies, and structural reforms designed to transform scientific productivity and technology regulation in the United States.

This conversation examines the White House strategy to revitalize American science and technology through artificial intelligence, regulatory overhauls, and funding reforms. It explores how government initiatives can accelerate innovation, modernize research institutions, and maintain global leadership against international competitors.

AI and scientific productivity. Kratsios argues that applying artificial intelligence to research is the most critical mission for the nation, aiming to dramatically accelerate discovery across disciplines. He points to the Genesis mission, which recently launched about $5 billion in grants to integrate AI into federal labs and double the productivity of the American scientific enterprise.

Reforming scientific funding. Kratsios warns that current research incentives and academic structures are outdated and broken. He notes that researchers spend 45% of their time on administrative grant work, which is why the administration is introducing longer five-year grants and experimental funding models to support young talent.

Global AI competition. Kratsios argues that the United States must lead the world in providing the best AI infrastructure and chips. He explains that the administration launched the American AI Exports Program to package turnkey American AI stacks for global customers, countering foreign competitors like China.

Regulatory reform and infrastructure. Kratsios argues that outdated regulations stifle technological progress in fields ranging from supersonic flight to data centers. He points out that the administration is changing federal rules to replace strict speed limits with noise limits for supersonic jets while requiring tech companies to cover all associated power costs for data centers.

America's Long-Term Vision for Space and Technology

Ray Kurzweil predicts we'll see as much progress in the next decade as we've seen in the last century, like going from the Ford Model T to Starship in 10 years. On top of that, we're on the edge of AGI, maybe in the next 3 years, and then ASI.

Governments have struggled with this for a long time. We must ensure the regulatory systems around these areas of tremendous growth exist in a fashion that doesn't get in the way of progress. Some technologies are born free, while others are born in captivity. For born free technologies, like the internet in the 1990s, the best thing the government can do is step back.

Government's Role in Regulating Emerging Technologies

A bipartisan bill passed in 1996, backed by Bill Clinton, allowed some of this to take hold. Similarly, the EU AI Act was finalized by the EU Commission before ChatGPT was invented, so it cannot apply to today's LLMs.

Some technologies are born in captivity, such as commercial drone operations and AI-powered medical diagnostics. Their benefits will not be realized by the American people unless the government affirmatively acts. If we are to see extraordinary progress toward AGI and ASI in three years, I wonder if people here in the White House understand the speed of that change.

We are trying our best to bring people along to this new velocity of change, best manifested in our Genesis mission. The president, Secretary of Energy, and I stated that the most important thing for the nation is applying artificial intelligence to scientific discovery. Other nations have already appointed AI ministers; the Emirates has driven 50% of government operations to AI, and in Malaysia, the president has an AI representation capable of speaking all languages.

Since the first Trump administration, when President Trump signed the first executive order on artificial intelligence in 2019, our view has been that AI will impact every agency, whether they are flying drones, performing AI-powered medical diagnostics, or working on financial services at the SEC. In the short term, I hope all our agencies can start using AI. Currently, large language models are not allowed on our systems here in the White House complex because of the Presidential Records Act, but hopefully we will change that soon.

Public Perception and PR Challenges of AI

The challenge right now is that 3/4 of Americans fear AI. 71% oppose data centers near their homes, a larger percentage than those who object to nuclear power plants in their neighborhood, which is insane. AI has a massive PR problem.

I think back to the first AI safety summit held by the UK government in Bletchley Park two years ago under Rishi, perhaps the year after ChatGPT came out. When I attended, the entire concept was wrapped around fear: what could go wrong, and how the smartest people and most important government leaders must come together to ensure these harms don't impact the world. This is an example of how the government narrative prior to President Trump has been so fixated on the negative impacts of this technology; of course people are scared.

AI in Healthcare and Public Benefits

If all people hear from politicians and industry is that there will be job losses, danger, and bio risk, we must look for areas where Americans connect with AI positively. By far, we have seen this most in the healthcare domain, where we can show the impact on how individuals and their families receive care. It kills me that the narrative in China is the flip of that, where 80% are pro AI, at least in the AI labs.

The government can serve as a convener to bring to the surface the great stories in this industry. AI is having a dramatic impact on manufacturing and hiring across the country, including the huge build-outs supporting the industry. Jensen was here talking about the supply chain supporting Nvidia; he said the order he put in to Corning for future chips is the largest in Corning's history. They are producing more than ever, lifting the nation's GDP in an extraordinary fashion. Factories are being built, people are being hired, and stuff is being produced; these are the things that can connect with Americans.

One of the biggest fears is AI-related job loss. It is confusing because some put forward data showing layoffs, while others feel we will see more job creation, as we have with every technology so far.

Personally, I am very optimistic about the long-term impact AI will have on jobs. Since the first Trump administration, the narrative focused on automation and fears of job loss, but that was never realized and employment has only increased. Still, it is something we must address, and we have implemented programs to do so.

As you mentioned, the data on what is actually happening is not very good. We are launching an initiative at the Department of Labor, called for in the AI action plan, to collect better data on AI's impact on the labor force. By collecting data from players who typically do not submit it, we can make the right decisions on where to provide better reskilling, retraining, and targeted labor assistance programs.

At the FIA Summit in Miami, we discussed a parachute program to incentivize companies performing AI-related layoffs to provide employees with AI upskilling. I think many folks these days assign blame to AI for layoffs they would have done anyway because it plays better in the press and their stock price goes up if they produce more revenue with fewer people.

Fear is an awful place to face the future from, especially at this speed of change and when people do not understand AI's implications on their lives. This year, we launched the Future of Vision X Prize, the world's largest film competition for creators to show a hopeful, positive vision of the future where technology and AI work together.

America's Big Bold Missions in Space and Science

Part of the challenge for me is that the majority of sci-fi films, like Terminator, Ex Machina, and Black Mirror, are dystopian. I think about this more in terms of national missions, which we'll discuss in Golden Age. The government needs to be opinionated about what is most important for the future of our nation.

Beginning with the Manhattan Project and going through Apollo, we had big ideas that motivated young people to go into science. We looked at a North Star of something people thought couldn't be done, but we did it. An example is the space-related efforts you recently discussed with Administrator Isaacson and Jared. We are putting man back on the moon in '28, building the first elements of a lunar base by '30, and putting a nuclear reactor in space by '28. Telling someone we'll put a nuclear reactor in space with enough propulsion power to send folks to Mars in the next year and a half sounds nuts.

Then there is the Genesis mission, our AI for science mission, where we want to double the productivity of the entire scientific enterprise in the United States over the next decade. Finally, the president directed through an executive order the creation of a scientifically relevant quantum computer by the end of his term. While the basic research done in quantum for so long is amazing, we are putting a stake in the ground to build this machine and have it do something.

I am most excited about pharmaceuticals. I believe the calculations you can run on particular molecules for drugs will be transformational.

Outside of our term, fusion energy is important to keep thinking about. People have said it is 10 or 50 years away forever, but the Department of Energy put out its latest national strategy a few months ago. Never in history have we had more private sector investment in this energy source; by my last count, there are 37 venture-backed fusion companies.

Jensen released the open secure AI alliance. As an administration, we believe the US must lead the world in both closed and open source models. The best thing for the country is a vibrant ecosystem on both sides of that coin to support any customer who wants to work on or use AI. We have a couple of extraordinarily well-funded startups pursuing the front. Mira Murati's release was amazing, and I understand we are waiting on reflection for their model later this year.

The ecosystem and tooling around open source continue to improve, and the US remains the default for those harnesses and tools. As expressed on the first page of our AI action plan last year, the US has to lead on open source. Honestly, the Chinese have very well-performing open source models. If you are a cash-strapped American entrepreneur bootstrapping your company, I cannot blame you for using the cheapest model out there.

US Leadership in AI and Global Competition

At this moment it's Chinese, but I think over time we'll be able to cultivate a pretty vibrant ecosystem here. The Financial Times reported that Beijing is exporting open source as an instrument of influence, providing the global south access to infrastructure and capabilities.

One of the stated goals in the AI action plan is that the world should build on America's AI and tech stack. While many people will use frontier AI, the vast majority of the world will use on-prem open-source models. To operationalize the AI action plan, we launched the American AI Exports Program. Our vision is that we have the best AI stack in the world, with the best chips from companies like Nvidia, AMD, and new entrants, the best models, and the best applications. For any government or private sector customer looking to build AI, there is nothing better than the American stack.

We launched this program at Commerce, but it has become a whole-of-government effort. We put out an RFP asking American companies to submit proposals on what an American stack looks like. We are now reviewing those proposals to create turnkey American AI stack options for the world, backed by government financing organizations like the Export-Import Bank and the Development Finance Corporation to make them economically viable. Countries likely don't want to choose seven different vendors to set up their stack.

Another appeal is that everyone wants our chips; we have the best by far. Between our export controls and other factors, particularly on EUV, our lead over the best Chinese chip continues to increase year-over-year. We will continue to have the best product as the foundation of the stack.

This AI export program was born from the frustration I had in the first administration with Huawei and our inability to counter Chinese actions on telecom. They had a good enough telecom stack run by Huawei, super subsidized by the PRC, and they proliferated it quickly. We want to avoid that now. We have a vibrant open source ecosystem, and our open source solutions, plus our code source, chips, and the great American applications everyone wants to use, will make our stack more competitive.

In the launch vehicle and satellite defense industries, we saw that restricting exports simply cultivates competition to build their own capabilities and compete with us. I believe one of the savviest and most impactful decisions our administration made was throttling their ability to make models competitive to ours.

It has been a priority of Xi Jinping to have a competitive semiconductor industry for a long time. That is why the EUV lithography export controls in 2019 were so critical; they are probably the most impactful export controls in U.S. history. Because of them, their ability to create a competitor to our leading-edge chip has been hindered.

We have a number of great US-based humanoid robot companies, but you can count them on two hands compared to China's 150-plus. China has been aggressively supporting them through funding, national events, and centers for robotic advancement. I believe robotics will be a critical manifestation of AI for a country's future success.

This week, we took dramatic action by limiting the importation of any non-US humanoid robot that hasn't already been shipped. We must build the supply chain muscle to ensure future security. We took similar action around UAS or drones last December, and the resulting investment in the drone supply chain has been dramatic. We hope to see the same here, as this industry needs a push because the Chinese are subsidizing and dumping on robotics.

Elon, a friend of the pod, told me a few months ago that we will see double-digit GDP growth in the next 18-24 months and triple-digit growth in five years. He also told The Economist that by 2036 we will be basically post-capitalist, where anything you could possibly want will be delivered by AI and robotics.

Challenges in Scientific Productivity and Innovation

Elon is the most brilliant engineer on the planet. Most agree his predictions are directionally correct, even if the timing is off, but the speed of change he projects breaks every system. While I may not ascribe to that particular velocity, our cabinet and White House leadership recognize things are changing and are getting in front of it. Secretary Bessette's involvement in post-mythos activities at USG is a great example; if a super capable cyber model in the hands of the wrong actor could risk our systemically important financial institutions, we must approach it seriously and quickly. That is the kind of action administration leaders are taking regarding these rapid changes.

I was a kid in a candy store reading the Golden Age report. Its first point is that scientific productivity has been declining despite larger budgets—Eroom's law, or Moore's law spelled backwards—meaning discovery per unit dollar has dropped despite better tools. We have been unable to change how we conduct science. After I was confirmed, the president wrote me a letter charging us to revitalize the science enterprise. We have better technology and larger budgets than ever, most notably in the biomedical field where the NIH budget has ballooned to almost $45 billion, yet drugs are more expensive than ever.

One conclusion we reached is that we are not experimenting enough in how we conduct science; we do the same thing over and over, putting in more money and believing outcomes will improve. The core thesis of the Golden Age report is that we must be more experimental and ambitious about testing new ideas. It is shocking that in the world of science, where you would expect people to be the most excited to try different methods, the community doesn't want anything to change. They want the system exactly as it was 30 years ago.

Reforming Scientific Funding and Experimentation

If you're an expert in something and there's a revolutionary breakthrough, you're no longer the expert. A well-known study by the National Academies found that researchers spend 45% of their time on administrative work associated with their grants; we want them doing their work 100% of the time.

One of my long-time pet projects is bringing supersonic flight back to America. This is a regulatory issue: the US has essentially had a speed limit for flights over land, meaning flying over Mach 1 is not allowed. Research and the company Boom Supersonic have shown they can fly over Mach 1 without creating a sonic boom, but current regulations disincentivize them. Due to a presidential executive order, we are changing that rule to create a noise limit rather than a speed limit. If you can keep it quiet, fly Mach 1.

The second challenge is young talent waiting too long to implement their bold ideas. I agree that our funding, publishing, and credit system was built for a world of human-paced discovery. Everyone is stuck needing to do research, get it published in a journal multiple times, and then seek tenure. The NIH is a bipartisan example of this. I heard a shocking statistic today that the median age of an intramural NIH scientist is 71, while the average age of a Nobel laureate's prize-winning work is in their mid-40s.

I am excited about four mechanisms for driving progress in the golden age of NIH, the first being long-duration, five-year grants funded on day one.

Over time, we've come to a comfort zone where most government grants are roughly 18 months, based on academic calendars and how quickly the government can review and adjudicate them. But that's not the pace of scientific discovery. Some endeavors are small experiments taking only three or four months, while others require longer periods for a brilliant scientist to explore an idea. Most scientists start working on their second grant before finishing the first just to keep the pace going.

At the core of the new golden age report, everything we do is in service of the scientist. Some want shorter duration grants, and some ideas need five years to play out. The government needs to make money available for those scientists making great breakthroughs, including proof-of-concept grants of a few pages reviewed in under a month. This came to the fore during COVID, when Tyler Cowen and others pooled capital to make grant decisions in hours. There's no reason this should be restricted to times of crisis, as answering certain ideas may unlock others later.

Many scientists are likely using ChatGPT or Gemini to write grants, and I expect grant reviews will also use AI. Our funding agencies are thinking about how to accelerate that while maintaining the merit-based review gold standard.

I also love the experimentation with golden tickets. Many say we're not risk-taking anymore, and the least common denominator ideas typically get funded. People with crazier, out-there ideas that may not work—but would be incredible if they did—aren't incentivized to apply because they won't get the award. With the golden ticket concept, each person on a merit review panel of three or four people gets a few golden tickets to unilaterally fund a grant independent of the committee. This incentivizes crazier ideas because you only need one person to believe in you, and it incentivizes better people to join review panels.

Innovative Funding and Incentive Models for Science

Rather than having mid-tier people do reviews, we can bring in even better people. I was talking to an extremely successful professor at Harvard who said he was getting dinged in his reviews because he'd gotten too many grants and was too successful, and they needed to give other people a chance. The whole peer review process is unfortunately extraordinarily broken.

Government needs to be a good steward of taxpayer dollars, so we must be thoughtful in how we develop and evaluate our programs. One piece of Golden Age is the concept of meta-science—the science of science—which is our ability to evaluate how we conduct science and funding, and course correcting when they aren't working. For example, we may believe fast track grants are a great idea, but we need to run the experiment, see which types work and which don't, and update the next iteration. In Golden Age, we call for launching meta-science units, which have already been announced at places like NIH and NSF.

Imagine a scientific marketplace where funders post bounties for breakthroughs. AI agents identify promising leads, hire autonomous labs, and verify cryptographically signed results. Agents exchange data, hypotheses, compute, and funding through microtransactions, while smart contracts release payments as milestones are met. Prediction markets could guide grant makers, bounty markets could direct resources toward unsolved problems, and reputation systems could identify reliable agents. This system would operate continuously at machine speed, replacing slow institutional coordination with market incentives, while human experts remain essential for judgment in deciding which scientific questions and breakthroughs matter most.

What you're describing is a fundamental AI-native agent reimagining of the entire scientific process. In the report, I suggest that incentives aren't always easily aligned in our current system, but over time we will have technical solutions to bring those incentives together and share information at a speed that allows DAOs and agents to run the cycle millions of times faster than a human. Philanthropic scientific capital is greater today than ever in human history; the OpenAI Foundation itself is almost a quarter trillion dollars in today's valuation. We have an opportunity for really smart people to push the envelope and try some of this stuff.

Folks like Yuri Milner, Eric Schmidt, and Marc Benioff are funding science directly. In 1950, when Vannevar Bush wrote Endless Frontier and kicked this all off, 70% of R&D was done by the federal government and 30% by the private sector. Now, the majority is done in the private sector and only about 30% is funded by the federal government, with philanthropy playing a bigger role. You can bring all those pieces on the chessboard together to drive scientific discovery in a way unimaginable in a system designed in 1960.

For example, we announced four-year PhDs today to get more PhDs out faster and prepare their experience for jobs in industry, not only academia. Having a private sector company paired with an academic institution to help a student pursue a four-year PhD is amazing and reflective of today's reality, not something you'd imagine in 1960.

I have spent 30 years of my life focused on incentive prizes. At X Prize, we've launched $600 million of prizes and driven about $30 billion of R&D. If there are big national scientific endeavors or problems we need to solve, we can pull money together to make prizes even bigger and draw people to solve these challenges. I told Elon, "Let's launch ten $1 billion prizes" to focus every grad student and company. The government spends about $200 billion a year in R&D; while a billion dollars may be a lot for the government budget to swallow for a single prize, $100 million is certainly doable if the project is big enough.

Our largest X Prize right now is the $101 million X Prize Healthspan. We have 830 teams competing to reverse functional age by 20 years, providing cognitive abilities, muscular, and immune system capabilities that are 20 years younger. I firmly believe, and scientists like David Sinclair and George Church agree, that this would impact the US economy most positively. Today in the United States, the average life expectancy is about 78 or 79, but the average health expectancy is 63, meaning the last 16 years of your life are in poor health.

The Future of AI, Robotics, and National Missions

Giving people an extra decade of health and productivity would transform everything. It is critically important to the general health of Americans, and we should find ways to work together to figure out what programs the NIH and other places can accelerate. Many believe aging is a disease that can be slowed, if not cured. Dario suggested doubling the human lifespan in the next 5 to 10 years, and Demis is dead serious that AI can cure all diseases within the next decade.

Looking at the six national technology missions—AI, quantum, commercial fusion, lunar exploration, robotics, and next generation semiconductors—I find biotech missing. Much of that work has been nested under the Genesis mission, the first national initiative around AI for science. We announced about $5 billion worth of Genesis mission grants last week, many of which relate to biotech.

Some states and cities are far more pro-tech than others. Max put together a map of what I call "singularity zones" where they are pro-autonomous vehicles, drones, nuclear, and data centers. China and Chile have done this to accelerate work and encourage entrepreneurs to move there and build their ecosystem. As an administration, I have been focused on this to drive innovation.

In 2017, the first paper I got the president to sign launched the UAS Innovation Pilot Program. At the time, commercial drone delivery wasn't really a thing, so the EO called for a pilot program at the Department of Transportation where the FAA picked 10 state, local, or tribal governments to pair with UAS operators for tests. We have done that again with our EIPP, our EVTOL program, and we will start seeing testing of EVTOL vehicles across the country. During the 2024 campaign, the president released videos about creating these innovation zones in cities to test new technologies. In our state-by-state evaluation, Texas came out number one in everything.

Transforming Education for the AI Era

Federalism is core to us as a country, allowing states to make decisions so people and companies can vote with their feet and go where they are most comfortable. For example, in the Elon conversation, we saw self-driving vehicles being tested in California leave for Arizona when the rules changed. This is regulatory arbitrage.

This dovetails with issues on data centers, where communities may compete for them in the future. Data center operators should think more about the incentives they provide so communities are interested in competing for them. The president has taken leadership on this with the ratepayer protection pledge, directing big tech, AI companies, and data center builders to build, bring, or buy all their own power. You are not to enter any community unless you cover all associated costs; if anything, you should lower electricity costs. Data shows that locations with AI data centers are getting lower electrical costs on average. Furthermore, golf courses use 30 times more water than data centers, and almond farming uses 50 or 60 times more.

As a dad of two 15-year-olds, I am pissed at the educational system today. In the US, it is still tied to the Industrial Revolution rather than what is coming. It should be AI first across the board because the public educational system is not serving our kids' future for the world that is racing at us, to use Elon's term, like a supersonic tsunami. We should be providing parents options for the ways they believe they can best educate their children.

There are amazing opportunities for people on one extreme to participate in things like Alpha school, where you do three or four hours of AI in the morning and spend the rest of the day on social skills and other subjects. Other parents prefer no technology in the classroom, wanting their kids to learn the classics and discover other technologies in different domains.

In many districts around the country, the majority of students graduating from high school can't do basic math. One of the most depressing things about American education today is the declining number of students wanting to enter STEM fields, which harms our country's health, national security, and future economic growth. I believe the government can inspire people to return to this domain. The Apollo program and Star Trek are 100% responsible for everything I've done in my life.

Asking school systems to change over the next few years—the time horizon we need—is extraordinarily difficult with teacher unions and public education boards. I wonder if there is an opportunity for an AI educational overlay available for parents and kids in the afternoons or weekends. The Alpha school founders started by running schools in certain locations, which isn't necessarily cheap, but the goal is to open source the software or make it economical for anyone to run those programs.

The Vision for Accelerated Scientific Discovery

AI proposes the hypothesis, robots run the experiment, and the model reads the results to design the next one 24/7, with no humans in the loop. You're running closed-loop science at machine speeds. Lyla Sciences is building out a million square feet, and we believe these should be retrofitted into federal labs using federal dollars.

A long pole in the tent is building hardware capable of running a wide range of experiments, so we'll likely start in narrow domains and expand as the hardware evolves. What has held scientists back is how long it takes to run experiments. Imagine if you could go online, enter a hypothesis, hit go, and start thinking about the next one.

I love the vision where a high school or college kid could get access to national lab equipment. In high school, a few kids always found ways to meet professors and run experiments in university labs. This democratizes the process, giving anyone with ideas the opportunity to test them without the overhead and burden of institutions that hold the keys to the infrastructure.

I spent ten years in college and graduate work designing experiments, pipetting, and plating dishes. Universities will likely be among the first to build these labs to provide the best tools to their scientists; administrators would do a massive disservice to their academic community if they didn't. These are the resources everyone will be demanding.

In the Genesis report, we talk about doubling productivity. In my moonshots world, 2x is low—we ask how to 10x it. That 2x number came from last year when we were building the program for the Genesis mission in December. Government is one of the hardest institutions to move; it's linear at best. Elon learned this firsthand and did God's work helping the country. We should be aiming for 10x, and given the transformations in AI over the last six months, we should definitely be pushing for it.

Early this year, Opus wasn't even out yet. Anthropic revenue was sub-10 billion, and it's now the fastest growing company on the planet. We also had the Mythos moment and the Fable release, and OpenAI is releasing their new 6 model very soon. We have to do 10x.

Alex Weisman, Gross, and I wrote a paper at solveeverything.org. We look at how to structure the situation such that AI is able to use, in Alex's words, "cooking math." Math as a discipline is being wholesale solved over and over again, with new breakthroughs and proofs being dismantled or proven. On the heels of that come physics, chemistry, biology, and material sciences.

We imagine an inflection point where GPT-6, whatever follows for Anthropic, and whatever Grok becomes—especially now that Elon uploaded all of SpaceX's engineering data, which was a baller move—accelerate the rate of science in a way that is stunning to people. We believe every scientific domain, whether chemistry, physics, math, or biology, will see a dramatic transformation and acceleration of discovery and hypothesis testing.

To unlock that, our insight was that US national labs have a tremendous amount of scientific data across a wide variety of domains that have not been made AI ready, uploaded to models, or part of the scientific process. We have a huge opportunity to bring 70 years of US scientific discovery into an AI model to accelerate discovery.

When that data, such as wholesale health data from the NIH, is brought in and breakthroughs occur, does the value inure to the American people? Who gets the financial benefit from breakthroughs coming out of federal data? Think of it like weather data; the federal government made all NOAA weather data freely available, so now your weather app and everything else essentially runs on NOAA data.

President Milei of Argentina has taken an aggressive stance, saying there will be no taxes on AI companies and enabling AI personhood for agents. For us, we want to focus on making sure the benefits of AI are realized by American people. We have the best companies in the world and need a regulatory environment that allows them to keep making the best breakthroughs in the world.

Mentioned

People

Ray Kurzweil Bill Clinton Mira Murati Xi Jinping Tyler Cowen Yuri Milner Eric Schmidt Marc Benioff Vannevar Bush David Sinclair George Church Alex AWG Alex Weisman Gross

Companies

NVIDIA Corning Department of Labor Department of Energy AMD Export-Import Bank Development Finance Corporation Huawei NIH National Academies Boom Supersonic Harvard NSF OpenAI Foundation X Prize FAA Lyla Sciences Anthropic SpaceX DOE