Winning the AI Race Part 1: Michael Kratsios, Kelly Loeffler, Shyam Sankar, Chris Power

5 4 3 2 1 zero. All engine running. Liftoff. We have a liftoff. That's one small step for man, one giant lift for mankind. The world's largest airliner. Each wing is big enough to hold five tennis courts. This new technology made it possible to meet the user's crucial needs. Enter the computer and a new age. What a computer is to me is it's the most remarkable tool that we've ever come up with. And it's the equivalent of a bicycle for our minds. Here I am playing a game of chess with a computer which is analyzing board positions and applying a certain kind of intelligence to figure out what its next move should be. That's the subject of our program today, artificial intelligence. The good future of AI is one of immense prosperity where there is an age of abundance. Everyone can have whatever they want. We're still in the very early innings of AI. I would say the rate of progress is exponential right now. Every time I think that we are overstating the impact of artificial intelligence, something comes along that tells me we aren't making enough of it on the show. You know, there's no 60-minute clock on this thing. This is an infinite game. Think about solving a problem that would take humans thousands of years to solve. Those who can harness and govern the things that are technologically superior will win and it will drive economic vibrancy and military supremacy. The Trump administration believes that AI will have countless revolutionary applications. We believe that America's destiny is to dominate every industry and be the first in every technology and that includes being the world's number one superpower in artificial intelligence. It feels like every tech revolution of our lifetime has been leading to this moment. All right, everybody. Welcome to Winning the AI Race. This is our first event in D.C.

Thanks for coming out, everybody. We put this event together in just a couple of weeks in order to have a really important discussion: winning the AI race. This is something America has to do, and it's something we will do. We're going to do it through the way we've won every other technological race: through grit, entrepreneurship, and dogged competition.

The difference with this administration is that it's actually engaging with the technology industry. Today we're bringing together many members—in fact, all members—of the administration to talk about it. None of this would have been possible without our bestie, David Sacks, deciding that he would take some time and become our czar of crypto and AI. I would like to start with a huge round of applause for David Sacks. David, you've been here for 6 months.

I'm sorry, but did you actually prepare? This is excellent so far.

No, it's excellent. Keep going. Keep going. He wants to be invited back to D.C. They told me I've got 12 hours left on the ground.

I think that White House tour is going to happen after all.

It might just happen. It might just happen in the air.

But in all seriousness, you've been here for 6 months, and we all know how capable you are. My lord, this administration is on a heater when it comes to crypto and AI. I am absolutely—and I think I speak for everybody in our industry—thankful and wildly impressed, but not surprised, at the pace at which you've led crypto and AI. What's the first 6 months been like for you?

It's really been incredible. I never expected to go into government at all, largely as a result of President Trump coming on our podcast about a year or so ago. That began a relationship that eventually led to my being offered this job, and I took it because I thought it was a once-in-a-lifetime opportunity to work for a president who really wants to get things done for the American people. You can see that he works so hard every day to push forward his agenda for the American people, and I think AI and crypto have just been 2 of those issues. It's been a lot of fun to work on these things because we are getting a lot done.

Yeah. Last week, you put this event together in just the last 10 days as an opportunity to talk about your—

The president's action plan was getting published today. But we should invite Jacob out, because Jacob partnered with us from Hill & Valley. Jacob Helberg, come on out and join us.

Yep. Our new fifth bestie, Jacob Helberg.

There you go.

Nice to see you, brother.

Good to see you.

Good to see you.

Good to see you. Friedberg, your team and Jacob's put a ton of work into this, and we have a lineup that is absolutely outstanding. So thank you, Jacob, for the hard work from your team, and Friedberg, for the hard work from your team to put this all together. Maybe you could give everybody an idea of the questions we want to address today and what the format is going to be.

Absolutely. The Hill & Valley Forum is a community of technology builders and policymakers who believe that technology is an engine of wealth creation and an indispensable pillar for American national security. It was incredibly exciting to have the opportunity to engage in this event, which is going to cover a lot of the topics that everyone in our community cares about. Ultimately, we believe—and I actually said this in my confirmation hearing not too long ago—that we're at an inflection point.

We are in an AI race, and the different parts of today's programming will be a series of conversations covering the different facets of how technology will actually create wealth for our country. I think it's very important, as we talked about who we wanted to have on stage and how we wanted to talk about the president's action plan that David shared with us, to highlight that there are new industries being created because of AI—industries that couldn't have existed a decade ago. We've got a couple of those examples, and then there are industries that are enabling and accelerating AI: mining, energy, chips, fabs, and data centers. We've got conversations across each of those.

That's this enabling conversation. Fortunately, we've been able to get a lot of folks from the administration to join us today to talk about the government's role in enabling this economic transformation that's already underway. I just want to say one point: I think what's become apparent to me, and I think is wrong in the press narrative today, is that AI is destroying jobs. I think what we are seeing on the ground is an incredible job-creation engine that's underway.

So I think it's very important to highlight that and share those stories, because they're not told enough. I think there's a real opportunity to bring them to light, and that's hopefully what we can get through today.

And Chamath, just coming around the horn here: it doesn't matter if you're a Democrat, Republican, independent, or moderate; this issue transcends party. This is the issue of our lifetime, and there are a lot of hard questions and a lot of hard debate. Maybe you could just speak to this administration's ability to bring in a lot of disparate opinions and work together across the aisle and with all members of the industry.

Look, historically, you've had a fork in the road where you can view technology as either optimistic and glass half full or pessimistic and glass half empty. The optimistic, glass-half-full view says that the country that can harness AI or any of these leading, critical-edge technologies is able to garner most of the gains. Then those economic gains can be spread.

It's a debate about how to spread those gains within a country and within an economy, and then from there, with economic supremacy, you also have military dominance. Now you're a superpower, and you remain strong. The problem is that historically we've gone in the other direction, where there has been this mistrust. In that mistrust, you've had global competitors emerge and create real, fundamental, existential risk for our place as a superpower.

Well said.

The great thing over these last, frankly, 6 months has been a massive pivot back into this idea that America is the best. We should not be ashamed of the things that we've created, and these incredible technologies and these incredible people should be celebrated.

Yeah.

Let's go and win the race.

Let's win. Okay.

And, by the way, how we're going to win the race—

Sacks, I know you invited Michael Kratsios to join us here today, the director of the Office of Science and Technology Policy. Should we have Michael come out?

Yes, Michael, come out.

Yeah.

Michael, come on out.

Please welcome Michael.

All right. How are you guys? Great to see you.

Thank you.

I'll kick this off. President Trump, in his first week in office, signed an executive order that directed us to create this action plan. Michael and I, along with the national security adviser, were tasked with figuring out how the United States would dominate in AI.

From his first week in office, President Trump has made this a priority, and we do see it as a global competition, or a global race. The consequences of losing that race would just be unthinkable, because AI is going to have such huge ramifications for our economy and also for our national security.

So, the United States has to win it. Working with Michael and the Office of Science and Technology Policy (OSTP), we put out a plan today that has 90 concrete actions that at least the executive branch can take to help us win the AI race.

I want to call on Michael in just 1 second, but I'm just going to outline the 3 big pillars of the plan. Number 1 is innovation. There's just no substitute for innovation. You have to out-innovate your global competition. You can't regulate your way to winning the AI race.

We have a lot of things in the plan that are going to help our private sector, our startups, and our tech community out-innovate the competition. Number 2 is infrastructure. We have to have more and better AI infrastructure—data centers, energy, and manufacturing—in the United States.

Number 3 is the AI ecosystem. We want to have the biggest ecosystem. We know from Silicon Valley that the companies that create the biggest ecosystems are the ones that win. You have the most developers on your platform, and you have the most apps in your app store. Those are the companies that dominate industries.

In a similar way, the United States has to dominate by creating the AI stack for the entire world. Those are the 3 big pillars of this plan. Let me call on Michael. Can you tell us how this plan was created over the last 6 months? I know your office did a ton of work on this, and then, if you want, flesh out some more of the important details as you see them.

Yeah, absolutely. Once the president signed the executive order assigning us this task, the first thing we did was issue an RFI—a very exciting government activity. We asked the country, “What should we include in this plan?” To be honest, I think we were all surprised by what came back.

We had over 10,000 responses from all corners of the country. We had Hollywood actors sending us responses. We obviously had tech companies. We had everyone you can imagine. I think it really showed how impactful this particular technology is to everyone in every industry in the United States.

We ingested a lot of those comments, went out to all the agencies that work with and, in some ways, touch AI, and came together with this plan. There have been a lot of national strategies that countries have put out over the last 5 or 6 years. What we really wanted to focus on—and it's in the title itself—was an action plan.

We wanted things that we could accomplish in the next 6 to 9 months to accelerate and ensure that we can win this race. If you think about the first pillar, which Jacob talked about, the innovation pillar, what's really key about innovation is that we want the next great AI discoveries to continue to happen here in the United States.

We have to create an environment that allows that to happen. When we talk about deregulation, the way I like to think about it is that there is never really going to be a law that says, “Hey, this is how we regulate AI.” What ultimately is going to happen is that these AI technologies are going to be built into so many other technologies, whether it's drones flying, self-driving cars, or FDA-approved AI-powered medical diagnostics.

All these different agencies are going to be touching technologies that are powered by AI, and it is incumbent on us to create a regulatory environment where these technologies can thrive and not be hindered by the government.

The next piece of innovation that I think is really key is using the power and the data that the government has to drive scientific discovery through artificial intelligence. We have seen great progress in this first wave of AI in the way that LLMs are able to handle coding, for example, but we can do so much more than that.

There are incredible data sets that the Department of Energy has, for example, at its national labs, that can help power a lot of next-generation discoveries in materials science and medicine. That's what this AI plan calls for and drives.

The next pillar is infrastructure. People talk about this all the time, and it's about how you create a regulatory environment that encourages and actually accelerates the ability of our power generators and our chip builders to do what they need to do here in the United States.

The plan calls for categorical exclusions for AI-related activities, which can allow data centers and other power generation to happen on federal lands. That's going to be coupled with all sorts of other efforts to really accelerate the velocity at which we can build power and ultimately run these data centers.

Let's talk, before we run out of time, about one of the most important issues, which is the talent wars.

Yeah, we're going to stay focused on AI here. We'll leave the border and deportations off the table, but we'll talk about something super important, which is recruiting talent from around the world.

This administration has gotten different signals, and obviously it's a very controversial issue here in the United States. What do we have to do in terms of immigration? Let's just call it recruitment, because that's really what it is: recruiting the best and brightest from around the world to come work on our team as opposed to, say, Team China.

What is the administration's philosophy on recruiting the world's best AI talent?

In the action plan, I think what we bring to light—and I don't think it's talked about enough—is that to power and successfully drive continued American leadership in this domain, it's not simply about having the greatest AI engineers. It's also about having all the other parts of the workforce that need to drive this forward.

We talked to some companies like Crusoe and others that are building these large infrastructure projects around the United States. The challenge that they're facing is a shortage of electricians and HVAC talent. The AI plan itself spends a lot of time and energy directing various agencies, whether it's the Department of Labor or others that have reskilling programs, to train these people up to be able to fill that void.

For us, it's about attracting the greatest scientists and engineers to the United States, but it's also about training the American workforce to be able to do the necessary jobs to put that forward.

Michael, what's the philosophy going forward on the thing you mentioned just before this, which is that there are these enormously valuable data sets that sit inside the Department of Energy and the FDA, where presumably, if we made them available to private industry—particularly American private industry—the gains could be incredible?

Is that an open-source philosophy? Is that a licensing philosophy? How do you think it should best serve the American economy to get this stuff out there?

Generally, the government has taken an open-source approach to this. The general challenge that we've seen over the years is that there's been a lot of lip service to, “Hey, let's unlock data for the American people.” The main challenge, for all of us who are in AI, is that the format of the data itself actually matters a lot.

If it's dirty, nasty data that isn't homogenized in any way, it's not particularly helpful. I think that's going to be a big effort that DOE—the Department of Energy—is going to try to undertake to make this better and possible.

What's great about the recent legislation that was passed, the One Big Beautiful Bill (BBB), is that it actually included a $150 million allocation to the Department of Energy to build an AI-for-science program that very much is going to be working on this exact problem.

Should there be federal preemption on AI regulatory schemes? There's been a conversation about doing this to ensure—I think right now there are over 1,000 state laws that have been proposed or passed that have some regulatory effect on AI and tech-related technology. Should the federal government preempt all of that and raise it up?

I think generally preemption is an issue that comes up very often broadly in technology. You had this issue with privacy for many years. What we're trying to focus on today, and what we talk about in the plan itself, are actions that the executive branch can take itself.

A lot of the preemption discussion revolves around what Congress can or can't do, so we don't necessarily lean hard on that because we focus on things we can accomplish.

Right.

And just to add to that, it's true that the action plan doesn't speak to that issue, frankly, very much. But I do think there is a real threat to national security brewing by virtue of the fact that, as you said, we've got 1,000 bills going through state legislatures right now, all regulating AI in different ways.

If this continues, we're going to have a patchwork of 50 different state regulatory regimes as opposed to 1 seamless national network. China has declared that AI is a national priority for them. They understand how strategic it is.

I think if we hobble our AI innovation with a patchwork of 50 different state regimes, it's going to hurt us. We weren't ready to declare a policy yet in the action plan, but I think it's something that's going to have to be looked at over the next year or so.

Thanks for joining us, Michael, the director of the Office of Science and Technology Policy at the White House. Well done.

Thanks.

Great job. Pleasure.

Thank you, everyone, to the Besties and the Hill & Valley Forum for the warm welcome. I'm Chris Power, the founder and CEO of Hadrian, and I'm here to talk to you today about our company. The mission is to reindustrialize America.

We do this by building AI-powered factories in the United States. You might ask why this is important and why you should care about manufacturing in the United States. What I realized before I came to this country is that we're in a global race.

Every great nation gets built by having the best industrial power first. That gives you the best military, usually the navy. Then you end up with the reserve currency after a conflict, and you kind of rule the free world in what we've called Pax Americana.

Like all great companies, you get lazy through that success, and you end up offshoring all your heavy industrials to developing countries. Then, when a conflict comes around, you're in real trouble because you offshored the thing that gave you the power in the first place, which is heavy industry.

The last 3 times this happened, it was a pretty good trade for the West. It went from the Dutch to the British to the American Empire when we won World War II. This time around, in this kind of 2-decade period where we're fighting the AI race and the climate, and settling the stars, it's really the United States versus the CCP.

Bear in mind that we won World War II not because we necessarily had a defense industrial base, but because we were the industrial powerhouse of the world. When there was a time of crisis, we had all our commercial manufacturing companies pivot to defense when we really needed them the most.

We had watchmakers making warship navigation equipment. Ford switched from building cars to building bombers. It was because of this industrial power that our tanks weren't so great, but we had tons of them. This is how we won.

Unfortunately, from the 1970s through the 2020s, we've basically hollowed out the middle of America and offshored every bit of manufacturing we possibly can. This started with Nixon opening up China and letting them into the WTO. They were the world's factory.

This was a huge strategic mistake. It completely hollowed out good jobs in America, as well as left us in a very strategically dangerous position in terms of our industrial power. While China deindustrialized us, they industrialized themselves, and they treated manufacturing not as economics but as a national security priority.

Now we're in this 20-year window, staring down the threat of Taiwan, and we're in real trouble. So, just how far behind China are we?

In munitions, China has automated factories that can produce 1,000 a year, whereas we run out of missiles in the first 7 days of any wargame conflict. Then we can't reproduce that ammunition for 3 years.

In shipbuilding, they're 200 times greater than us. We produced a grand total of 5 ships last year. Pharmaceuticals are all offshore. Drones and iPhones—we don't make any of them.

Bear in mind that, in pharmaceuticals, the CCP makes all our antibiotics. This is why industrialization is so important. More importantly, this gets back to the AI race for talent: While the United States is still the global powerhouse in software and AI talent, we made China into the global powerhouse for manufacturing talent.

What we realized through building this company is that while U.S. defense manufacturing—which was all we have left because we offshored everything else—is really important, we let all those jobs go. The entire base is basically made up of patriotic Americans who still know how to do skilled trades, are in their 60s, and are retiring faster and faster.

The underpinnings of our entire defense industrial base are these American workers who know how to do the job, but the rest of the country forgot how to manufacture. This is a screenshot of one of China's munitions factories. You can Google this online.

It's a myth that it's just low-cost labor in China anymore. They are very advanced at production. Whereas in the United States, underpinning all our defense primes and our industrial base, we basically have skilled Americans who are retiring faster and faster, supporting $100–$200 billion industries across all these different ways to bend, cut, and shape metal that you need to then put into drones, ships, satellites, and rockets.

While China is racing ahead of us, we're really falling far behind, and we forgot how to manufacture. So what we realized was that we have to build full-stack AI-powered factories to solve this problem.

Secondly, the number-one issue is this massive skilled-talent shortage. Remember, if you look at shipbuilding or any of these other industries, we are begging for millions and millions of welders or machinists. You could give me $1 billion, and we can't hire them in this country anymore because we lost that skill.

Production, not having inventory, is real deterrence, and you've got to do this by reindustrializing the country to create more jobs, not replace them or automate them away. It's always about national security, not economics.

We set out to solve this problem by building automated factories driven by AI in the United States 3 years ago. When we started this journey, we figured out how we were going to do this. The answer was to start running a factory and build all the AI software at the same time, which was a hilariously painful journey in the early days of the company.

This is what Factory 1 looked like. We partnered with some of America's greatest aerospace companies to beta-test this for a good 18 months. What can we automate? What can't we?

This is one of our first tiny parts that we ship to America's greatest rocket provider. Now we're up to the point where we're building whole products. We built Opus, which is a full-stack platform for AI autonomy of factories that does a couple of really important things.

In 2024, we launched Factory 2. Once we were out of this beta phase, we scaled 10x in a single year with the fastest-growing manufacturer in the country, and now we're lucky enough to support America's greatest companies, both startups and defense primes.

This is what the most advanced factory, in our opinion, looks like in America today. This is our scaled Factory 2 in LA. Here you see metal coming from raw material, being shaved down into components with micron-precision tolerances that go on rockets, satellites, jets, and drones.

What you see as you go through this is that, in legacy industry, in a deindustrialized nation, you've got really skilled people on every machine. Hadrian's advanced factories look and operate more like a data center. We're really proud of having pulled this off, but the journey is not over yet because, again, this is a whole-of-nation, $100–$200 billion problem.

Where do we actually get to? What sort of productivity gains can you get in AI and manufacturing, and are we creating more jobs?

Firstly, most factories in the United States run at only 20% uptime. It's not really that productive. We have a 4x jump in manufacturing productivity. Secondly, and more importantly, we have a 10x jump in workforce productivity.

Again, because we have such a scarcity of skilled talent in this country, you actually need that AI-powered jump to create the capacity in this nation to be able to build ships, drones, and rockets. The second important thing is the speed of getting people into these jobs.

If you're an advanced manufacturer, it can take you up to a decade to get really good at what you do. Whereas at Hadrian, we've managed to make it so that we can train anyone in 30 days. Most importantly, 100% of our workforce are from non-factory backgrounds.

They've never set foot inside a factory before. These are folks straight out of high school, retired from the military, or people who had a desk job. They were a nurse or a bus driver, from 18 up to 40.

This is the most important thing that people have got to realize about the power of advanced AI in manufacturing: We need this productivity boost just to be able to compete with China and catch up on these skilled trades that we lost. This is the most important thing that AI is doing for us: enabling huge, huge workforce growth.

So where are we at? In 2025, we're going multicategory, multifactory, and I'll show you our new factory that's launching AI-powered in 6 months in the great state of Arizona, as well as launching a dedicated gigafactory.

You can think about this as everyone in defense and aerospace needing a Tesla Model 3 factory. This is our beautiful new facility. It's about 4 times the size of the one in LA, launching by Christmas. We signed a lease a couple of weeks ago, and it'll be online in 6 months.

The most important thing is that we'll be creating 350-plus new AI-powered jobs in scarce-talent industries where America needs this leverage to get ahead. The other thing is, if you listen to the Secretary of the Navy and ask what the number-one problem is in shipbuilding, submarine bases, and munitions, it's actually that there are millions and millions of jobs that we need to fill because we don't have skilled trades anymore.

We don't have the volume of people, so we need this productivity boost. In 2026, we're launching advanced factories targeted at America's greatest production challenges: submarines, ships, and munitions.

By the end of this year, we'll have 3 facilities up and running: our headquarters, Factory 2, and Factory 3 in LA. But as we reindustrialize the country powered by AI, where is this really going to get us?

To solve this problem for the country and fulfill the mission, we need to have factories in every state. You've got to remember that AI in manufacturing is creating thousands of jobs because we offshored everything. We need this productivity boost to give our nation the capacity it needs, reshore all these jobs, pull them back into the middle of the country, and make sure that we're creating millions and millions of jobs along the way.

Thank you for having me. It was a pleasure to be here.

Chris, I think we wanted to kick this off. We have a couple of minutes to cover what you've introduced, which is, I think, a really important opportunity. China has roughly 3 million factories; the U.S. has 250,000. The assumption is they've got cheap labor, but it looks like they've got automation. Things are very different on the ground than what folks read about as we try to compete. What industries are going to be first from a manufacturing perspective that we can actually compete in successfully, and do we need trade tariffs in order to succeed on the competitive landscape?

I think there are 2 really important points. One is that there are industries that we have to reshore. Specifically, in defense, we have to produce submarines, ships, and munitions. We have to produce things like rare-earth magnets and drones. We just have to do it.

The tariffs really help, and this trade policy is really important because you've got to understand that, yes, China is more competitive than us, but the CCP also nationally subsidizes the cost of energy and the cost of raw materials. Because we've degraded this capacity, like not having nuclear in the U.S., we can't compete on those raw inputs. So it'll start with our most critical industries first, but I think as AI goes through manufacturing, you'll create millions of jobs, and that will allow us to reshore more commercial volume, not just in defense. I think that's the most important thing.

You've talked about this degrading infrastructure and what that means in terms of the workforce, but reshoring also requires upskilling. I know you guys have this associate named Owen that you took, I think, literally straight out of Home Depot. Can you give us a little bit of his story and what that represents in terms of you guys upskilling labor?

It's really incredible. As I said in the presentation, 100% of our people have never set foot inside a factory before. I think we really didn't do a great job as a nation by convincing everyone they needed a 4-year college degree to have a really good job. We've hired people who were packing shelves at Home Depot, and now they're running 10 machines at once.

Actually, what we are seeing is that most of those people, when they're exposed to software or AI, are very smart. We've promoted a lot of those people into leadership, management, or software engineering roles. I think reindustrialization with AI is about creating new jobs, but also reattaching people to the Silicon Valley economy, not just having it on the coast and in the cities.

How are you going to compete with people doing gig work and making $30 or $40 an hour as DoorDash drivers? We have the lowest unemployment in our lifetimes—4% right now. Is it realistic to find all this labor out there, or do we have to have some people immigrate to this country in order to fill those jobs?

For us specifically in defense, we can't—we have no choice on immigration because it's a regulated environment. So we have to upskill Americans.

Secondly, what we see, maybe not in L.A. or the coastal cities but across the country, is lots of underemployment. Some of our favorite people have desk jobs where they're a paralegal, filling out forms, and they hate it. They want to come into factories and work on the national mission. I think for us it's a lot about getting people inspired.

Secondly, with this level of productivity jump, we can actually afford to give people incredibly good healthcare and incredibly good pay. I think a lot of Americans want to go back to work in a real environment that's for the national mission.

You showed some incredible images and video of these very intricate machines. Do you make the machines that then make all the machines, or is there a supply-chain risk?

There is a huge supply-chain risk. We actually invented, via the Air Force, a lot of these advanced machines, and we forgot how to make them. The main sources of supply are actually our allies. China is number 1, but we don't buy from them because they've got cybersecurity holes all over the place. Germany, South Korea, and Japan are the other sources.

The insight that we had was that they're actually just really dumb computers, and software and AI can upscale and overpower them and really give us a lead. But it is a huge supply-chain risk that we don't build the machines that build the machines in the country anymore.

Right. To economically compete, though, do you—I was trying to parse whether you were asking for the government to give you support, since the Chinese government is underwriting its companies with free energy. Are you explicitly asking the government to help with, say, paying for reskilling training or maybe, in some way, defraying your energy costs, or can you make this economically work?

We make it economically work because, in the U.S., there are really 2 markets. There's the stuff that has to be onshore for defense and aerospace, and then there's this offshore market that's 10 times larger. Commercial aircraft, for example—a lot of that is in China.

For us, we can compete in the U.S. because we've got to create all these new advanced jobs, because we just don't have the skills anymore. If we want to reshore the commercial volume that is not regulated to be onshore, we have to do tariffs and economic policy because it's not an even playing field. It is, right now, companies versus the CCP.

What would that look like in terms of execution? You would want them to pick up the retraining, the energy costs, and part of their salaries?

It's really three things. It's the cost of energy. It's the cost of raw materials, aluminum, and steel. 90% of the cost of that is actually energy. If we level that playing field, then we can go compete in what we're great at, which is the American software and the American spirit and AI-powered workforce.

So the silver bullet is energy.

And then tell us about the actual software. Do you have a team that's writing a lot of control systems and/or AI models themselves, or are you taking things that are off the shelf and fine-tuning them? How are you doing it?

Unfortunately, because American manufacturing software is 30 years behind Silicon Valley, we had to build everything ourselves, from scheduling systems to the deep tech. The key insight that we had is that the faster we grow, the more data we are labeling, right? So we always do things 80% automated, with a human in the loop.

As we label this complex manufacturing data, this is where our AI models actually kick in, because manufacturing has been offline for 30 years. There is no Stack Overflow. There's no GitHub codebase to train a model on. We had to train it ourselves off our own labeled data as our experts were ticking and tying the automation.

Traditional automation is purpose-built; it does one thing. A lot of engineering goes into making it do that one thing really well. Are you leveraging things like—to Shyam's question—vision-language-action models that allow you more extensibility with one particular piece of machinery? When does that start to happen from a tech perspective, in your view?

Right from the start. The odd way that customers translate data to their supply chain is by giving them 20-page PDFs full of hieroglyphics. We actually have to train huge vision models to interpret that. What does that mean? It's very complicated, and it usually takes an expert 50 hours to pore over it.

So it's vision models, training agents on the data, and also training agents with reinforcement learning: We made a part with automation—was it high quality or not? Did it actually work? Embedding all of these in the workflow in real time is the magic trick here with AI.

You're not doing this stuff just on the coast, right? Your next factory is more in middle America. How do you end up choosing where to put that?

The most important reason why we selected Arizona was permitting, energy, and regulations. We've got to go fast, right? We've got to build this in 6 months, and then we will expand into the middle of the country, left to right on the map.

I think the most important thing is that we're going to be able to expand into all these cities and states where the manufacturing jobs were destroyed, and we're going to bring them back.

Are you guys investors?

Oh, yeah. I just led the Series C, which we announced last week, and joined the board, much to Chris's chagrin.

You're on his board?

I'm on his board. Yeah. Is that terrifying?

It is very terrifying.

How long have you guys known each other?

Too long. Yeah, too long. I was a board observer for a while and tried to avoid getting the official seat.

We dated for a while, and then we got married.

Well, Chris, thanks for being here today.

Thanks for hosting me, guys. Pleasure.

Yeah, thanks for the education.

Thanks, man.

A company that started in my college dorm is now a company that manages over 500,000 of the world's most critical pieces of infrastructure. At Gecko Robotics, we build robots and AI models to help unlock the physical world. When we build robots, we want them to be able to fly, swim, crawl, and walk on any surface to gather the most amazing information and datasets that have been forgotten about in the physical data layers. All those data layers are incredibly valuable when you're able to unlock them and use AI models to drive incredible and important outcomes.

I started the company in the energy sector, deploying the technology to help prevent catastrophic failures in power plants. Now we've expanded into mining, metals, manufacturing, and defense. We're helping to deter conflict by getting ships out of dry dock on time and patrolling the borders. We're also helping the Air Force ensure that planes are in the air and not in hangars.

Just last week, when the president was in Pittsburgh, my hometown, we signed an amazing deal that ensures we can help revitalize manufacturing in the United States by helping to build ships and submarines. The energy sector has been incredible, and we're in a lot of other sectors as well. But what I've begun to realize is that the most impactful thing Gecko Robotics can do to help deter conflict—and our most impactful work for national security—is actually in the energy sector.

President Trump is absolutely right, and his executive order today calls out an extremely important reality: The companies that can unlock energy are going to be the ones that can dominate in the AI race. However, as you can see from the graph here, China is on pace by 2030 to have 3 times the amount of generation of the United States.

But this isn't the whole story. We constantly think about AI as an energy consumer. However, I'm here to tell you that artificial intelligence can actually be used to unlock energy production in ways that you've never seen before.

Inputs really matter to being able to unlock this potential. Every CEO I talk to in the energy, mining, manufacturing, and defense sectors will tell you that they're trying to figure out how to unlock artificial intelligence to supercharge everything. However, the value just isn't there.

It's no wonder. The consistent common factor between each one of these sectors is Joe. Joe is out there gathering information by hand, trying to diagnose problems and get physical data to drive really impactful decisions. But it's important to understand that Silicon Valley artificial intelligence researchers and software engineers can't do much with datasets coming off the backs of Joe, and Joe has been armed with the same technology for the past century.

So it's no wonder that impact isn't being unlocked in these sectors. Unfortunately for Joe, it's a very dangerous job as well. Someone dying while doing this job was actually one of the things that inspired me to build Gecko. We have to give Joe better tools in the new century.

What I'm going to walk you through right now is an example of exactly how we do that for the power sector. We send in robots—robots that gather information and datasets about the physical environment. In this case, a natural-gas power plant. We're understanding what the physical environment looks like, and then we send in other robots, like this dog over here.

The robot dog is gathering operational datasets to help supercharge Cantilever, our AI-powered platform where all the datasets come together. We sell an operations platform, and the datasets gathered in the physical world are what enable it. We also send in wall-climbing robots, and you can see them to your left and your right.

These robots go into physical environments and gather health data, all while the power plant is online. The health data is really important because we have to process it to be able to optimize and feed it into AI models. But again, this dataset just never existed before, so we had to go out and get it physically in the real world.

Robots like this supercharge our ability to drive models and create the largest efficiency gains. This power plant, for example, is supposed to be operating at 620 megawatts, but it's not reaching its capacity. It's only operating at 580. So how do you unlock that?

When you have all this information and these datasets that we've captured with robots, plus all the datasets that customers have, you're actually able to drive optimization and see how to impact efficiency and production. This AI model is looking at the datasets from the robot dogs, as well as the health data from the robots, to pinpoint that there's actually a steam issue going into the turbine.

Our ability to fix these things has unlocked a 1% improvement in efficiency for this site and many others that we work on. And this is just the first place that we looked.

It's also important to understand that the assets that power the grid are failing at a really fast rate. This power plant had assets like this tank that were decaying at an incredible rate. It was supposed to be reaching retirement pretty soon, but we were able to predictively determine how to extend the useful life of this asset by 10, 20, and 30 years from all this data.

When you accumulate all the kinds of impacts that you can have from this kind of technology, you get efficiency gains like this across the dozens of power plants we've been able to work on. If you extrapolate that across the thermal fleet in the United States, it would give you 11.9 gigawatts of new power without putting a shovel in the ground.

Energy can be unlocked using artificial intelligence. It's really important to understand this statement: AI shouldn't just consume energy; it should create energy. And that's what we're showing here.

Not to freak anybody out, but the Department of Energy just came out with a study that showed we have about 4 years of useful life left in the assets that power our grid. At this rate, there are going to be 100 times the number of blackouts by 2030 if we don't reverse this trend.

What we're able to show—not just with power plants, but with mining, metals, manufacturing, and defense assets as well—is that you can extend the useful life of infrastructure, in some cases by 30 years. On average, it's been about 35 years. This is extremely important to ensuring that we're able to reverse that trend and ensure that America is well positioned to lead in the energy race and enable and unlock artificial intelligence.

Let me summarize this. We've spent so much time—and I think J.D. Vance has done a great job highlighting how much effort and how many datasets have been gathered to power AI models in the digital world. That's what makes ChatGPT so addictive.

But remember, the physical world has been forgotten about. Our robots are going into the fog of war to try to decipher and unlock massive amounts of information and datasets that give America and our allies unfair advantages—unfair advantages to unlock things that we didn't even realize were there.

If you build software with an ontology based on first principles, gathering the data and building the software up from there, you're actually able to deliver impactful things for Joe, turning Joe into a Ph.D. scientist or engineer instead of forgetting about him, like a lot of Silicon Valley companies have in the past.

Unlocking potential physical-intelligence data drives artificial intelligence, and that's how you're going to win the AI race.

Thank you.

My name is Laura Deirdinas, and I'm a registered nurse here at Tampa General Hospital. For the last 17 years, I've served in the neurointensive care unit, where we care for the most vulnerable and critical-care patients.

Before utilizing AI, it would take hours to gather information by looking through chart reviews and talking to nurses and physicians. We relied on paper and pencil—a lot of paper stapled together, sometimes with outdated data by the time I was done going through 32 patients. This is how we would try to give reports.

Bringing in AI has significantly changed the culture on the unit. I had a charge nurse who never gave a multidisciplinary round or report. She came on board and said, “This is an amazing tool. Look at this. It has all my information already gathered and collected.” She was able to report out on the patients.

It was completely user-friendly. She said, “Laura, what is this?” It's creating excitement throughout the nursing community. Using AI has provided more time to be with you or your loved one at the bedside, where nurses should be. We are the heart of health care.

Matt Troutman. I'm the vice president and general manager for PRL Industries, a supplier of components for nuclear submarines whose components our servicemen and women depend on. We are a fully integrated foundry, pouring metal all the way through finished machined components.

2 months ago, we weren't getting after any of the problems on the shop floor. The engineering director told me all his team was doing was quoting a 3-day process to quote a part: paper files, old archives, data tables, emails, side communications—all of this ended up getting lost in the fray.

Now, using an AI tool, they're getting halfway through that process in minutes. It frees them up to get back out on the floor and do what an engineer does best: solve problems and provide the Navy with the best-quality products in the shortest amount of time.

Understanding part location and status is a game changer. We can now talk very clearly with the customer about whether that part needs to become the primary focus of the business because it's a critically needed part for ship construction. You get notified. It's an automatic notification. We can see the exact status. Here is the impact. How can we be better? How can we do more? This is how we're answering that call. With AI, we match the speed of the quality-management process to the speed of the workforce and the machine capabilities, and we will truly see a multi-step change in the amount of product that can come out of any company in this supply chain. More jobs for American workers here at PRL.

My name is Julie Nordberg. I'm a registered nurse leader here at UP Health System Marquette in the heart of Michigan's Upper Peninsula, and we are really the only game in town, as we like to say. The next closest hospital to us that could service us is downstate, which is about a 4-hour drive. Prior to using AI, it took a lot of time to go through the patients' charts to see where they needed to be. It took a lot of time just to try to communicate with people. I think that's a fear that everybody has, that AI is going to replace people. But AI in the way it's being used here could never replace our frontline staff. The vibe is one of excitement. Everybody's proud to be part of this and to say that we're doing it here and we're honing it in and tweaking it and using it to enhance our care and using it to help our staff. Having this kind of communication hub and facility snapshot has helped everybody. For the nursing staff, being able to see everything in one spot has revolutionized how they are able to provide care. I don't think anybody is sad to get rid of a meeting. The impact on patients is earlier detection, which means earlier treatment, which is a better outcome, life-saving for some of them. We don't have as much manpower as those big academic centers, so having the AI in the background doing some of that legwork for us is huge.

I joined PCNA in 2018. We have built over 11 billion batteries in the last 8 years. I walked out onto their massive production floor for the first time. I knew right then and there I wanted to make this technology accessible for anyone who wanted to learn it. People are coming from the tourism industry and the hospitality industry. Quite a few technicians fixed slot machines in a past life. People from automotive companies, people who are used to repairing cars, however, have never seen equipment at this scale and with this complexity. We don't really have to pick and choose what people's backgrounds are because we do have this very powerful learning tool that makes it easy for anyone to enter this industry. It is taking our historical maintenance records, pairing them with our machine data, and is now starting to understand early warning signs of a breakdown and deploy our technicians to equipment before it ever actually breaks. This helps minimize our production losses and keep our technicians safer. We're taking reactive events and turning them into predictive events. We used to honestly lose a lot of technicians because they would lose their confidence and think, ‘Hey, maybe this isn't for me.’ I pulled the supervisor off the floor and said, ‘Hey, you have to come listen to this idea and help us make it better because you're the one who lives it every day.’ They immediately started suggesting new features. They were telling us what was wrong with the old systems, and we were coming up with solutions on the spot. So this is really helping people feel like they belong here. We don't believe AI should replace human talent. We believe it should elevate it. Our workers are very excited. They have a tool that they can turn to to help them learn at their own pace. It really puts the power back into their hands.

All right, Chris Power has joined us from Hadrian, along with Jake Loosararian from Gecko Robotics. And Shyam, welcome.

Thank you. Great to be here.

Chris, you want to kick us off?

Yeah, thanks for having me. It’s nice to be here. I’m definitely, for the first time, feeling like a guest of the besties.

Don’t mess it up.

Yeah, this is great. So, Shyam, thanks for coming. We were talking a little bit earlier, and maybe this is a great place to start. Obviously, we have the good fortune to be investors in Palantir for 15 years. We’ve seen the growth of the company, but particularly lately, you’ve been pushing this messaging that AI is not a force for job destruction. It’s a force for job creation.

It’s also a way that you can give superpowers to the average American worker. Obviously, we’ve seen a little bit of content here and how it’s already doing that today.

I want to start by saying many of the workers in the video are actually here today joining us. Laura, the nurse from Tampa General, actually brought her 12-year-old daughter. I think the ultimate litmus test is not just how excited American workers are to leverage AI, but how excited they are for their children to exist in an America that’s really embraced AI.

Julie has 4 kids, and she would tell you how much this has not only transformed her view of her job, but also her view of her children’s future. I think the right frame here really is: How do we give the American worker superpowers?

We should not be aspiring to build things that make them 50% more efficient. They’re really 50 times more productive, and we should use that as our asymmetry in the competition here. Our strengths are not only AI, which is clearly an American phenomenon, but also the ingenuity of the American worker.

If you spend time on the factory floor, on the front line, you see a very different narrative emerging, where people are actually excited about these tools. Every single one of those workers, to a T, said AI is giving them more time to do what they do best: to spend time with the patient delivering care, to actually build the parts as an engineer, to solve the problems, and not to be caught up in all the coordination and paperwork that surrounds these things.

That’s the future we should be unleashing.

Can you generalize the adoption curve? What is it about a particular industry or use case that makes it an early adopter versus a middle or late adopter, now that you’re touching all these different industries? You probably have a good point of view on this.

Yeah, my take is actually a different dimension of slicing that. Where does the institution liberate its workers to drive adoption, versus where are they trying to force-fit some sort of solution top-down?

AI is a method of unleashing the agency of the worker and the creativity of the individual, and they’re the ones coming up with these use cases. Chris was talking about it from Hadrian, where you’d be surprised at how people with deep mechanical intuition, traditionally considered blue-collar workers, are the ones who are able to pick up the skills, build the applications, innovate on their own processes, and have that spread through the organization.

Mm-hmm. And are you seeing that you have to build vertical tools or generalized tools for some horizontal kind of set of users somewhere in the organization?

Well, I think that the opportunity with AI is really that you can unleash what’s different about your business from all the others. There’s a degree to which you can have generalized solutions, but there’s a lot of alpha to be captured by understanding what’s unique about how we do things.

How do we leverage human taste? Everyone is afraid of AI replacing the human. That’s not what I’m seeing. I’m seeing it make the person with the greatest taste more valuable, with an ability to spread that to the breadth of the organization.

Let’s talk about something beyond taste, which is also knowledge and skill, and tell us about AI inside of healthcare. I think that a lot of people probably think that we have an incredibly cutting-edge system of tools and software that helps doctors and nurses actually provide great care. What’s the actual reality that you guys are seeing?

Well, sadly, I think with the forced adoption of EHRs, what we saw is roughly a halving in the productivity of how many patients you can see per hour.

A halving?

A halving. Yes. So, we became half as productive. The opportunity is to work backward from the care that needs to be delivered. How do we build the tools around that? How do we help the nurses and the care staff spend more time with the patients and less time with the computer?

Do you guys see a world where, in order to facilitate that end market versus a different end market, you have an ensemble of many, many different techniques and approaches in AI? Or do you think it all sort of gets form-fit into this 1-trillion-parameter, huge, ginormous thing that kind of tries to do everything?

I think the cardinality of agents and models is very high. I think there will always be alpha to be achieved—improved differentiation, improved outcomes—by specializing to the use case. Now, it’s great to start with the general models, but you will specialize over time.

And do you feel pressure to do that now, or do you think that’ll just be a natural evolution over time?

Yeah, I think it’s a journey that people get on. You realize, “Wow, look how much better things have gotten with this. Now, how do I go get the next incremental piece of performance out of it?”

You know, I’m just having this thought as we sit here and discuss this. If you think about any experience we have in a service that has a long wait time, where we feel like we got more time with the practitioner, it’s the perfect place for AI to create more abundance.

Healthcare and education are the 2 that come to mind, where people could just offload their chores, and the people who are getting the service can use AI to maybe start the conversation on second base or third base.

What other industries are you seeing after healthcare and education where AI can have that dramatic of an effect? The 6-week wait time to see a doctor, the 3 or 4 other students who are getting tutored are ahead of you, and maybe you don't need as much help, so you never get the tutoring.

The place I'm most excited about is really reindustrialization, because there's so much dwell time in the value chains.

What does that mean—dwell time in industry?

You're just waiting for someone else to figure out how to approve something, or the coordination costs mean that it's essentially deadweight loss.

Give an example there.

You saw it with the submarine industrial-base partners, where they're working on quoting a part for the Navy. That means you have to go gather all of this data and look at historical archives. All of that takes time. You're not making a part or solving problems; it's just sitting there. The factory floor is idle, right?

So how do we get rid of that dwell time so that you can utilize the capex that you actually have to the maximum extent possible? And then, if you zoom out, that's 1 part manufacturer. You exist in a massively complicated supply chain, and you just end up with all these busy waits along the way.

Yeah, that's so profound. A friend of mine who's in that industry said, "You're only as efficient as your worst supplier."

Exactly. And a second part of that, which the Panasonic Energy example really touched on, is how we train our workers. Here you have exquisite Japanese technology. It used to take 3 years to train a worker on it. Now, with an AI assistant, workers who were previously casino workers—they're not from this industry—are able to get up the curve in 3 months.

So you think about how we can use that to more quickly absorb the slack that's happening as we adopt AI and democratize opportunities. I have so much conviction in this that we've launched the American Tech Fellows program at Palantir to find blue-collar workers at our customers in the heartland—overlooked folks who have a natural proclivity.

How do you find them? Beyond just saying, "Apply," how do you find them?

Yeah, some of them are at our current customers. The idea really came from us organically. It's like, wow, who is building the most compelling applications? It's the guy on the factory floor, not a formally credentialed computer scientist—mostly an autodidact—but there's immense not only grit, but ambition.

They have the drive to reshape their own organization and reshape the processes. Let's bet on that person.

Going back to an earlier point, does that mean—and I'll ask the same question many times today—that college education, the traditional 4-year liberal arts degree, doesn't matter as much? That kids can go from high school, or earlier in their careers, into a new workforce and get well trained and well suited to make money and succeed in life?

Yes. I think the traditional college degree is dead, and we should be betting on the American worker.

Well, on that point, can you talk about the Tech Fellowship? I recently got to see a bunch of demos from the first cohort with you, and it's really incredible what you guys are doing there. Maybe give a little bit there, and then maybe also talk about the opportunity for other companies to follow this trade-school framework as we end here.

Yeah, it's really kind of an elite trade school. We're finding people with mechanical intuition who have done things. Some of them are right out of college; some of them have 20 years of experience.

This is the first trade school that you guys have done, right?

Yeah, that's right. And we have just enormous demand from our customers. We're asking, who are the people who have these skills? They're not classically trained, college-educated people. They don't have these skills, actually.

So the market's not meeting that need, and companies don't know how to source these folks. I can credential them, put them through the boot camp in 4 weeks, and place them with my customers to go unleash AI within their organizations.

It's incredible.

Paul Buchheit, you created Gmail, talking about efficiency and making the world more efficient. I believe we worked together.

You came up with the slogan "Don't be evil."

We worked together, too.

The slogan "Don't be evil."

Yes.

How'd that turn out?

It's an attempt at alignment, right? We worry about AI alignment. What do you tell the super AI once you've built it?

Yeah. You're at Y Combinator now. Although you recently said you're stepping down, right? Or you're—

Partner emeritus. We're starting a new firm, Standard Capital.

Oh, that's exciting.

Yeah.

Let's talk about the game on the field with startups. You get to see startups in year 0 and year 1. One of the primary theses I think we all have is vibe coding and making coding not a roadblock.

I think Paul Graham's great innovation at Y Combinator was saying, "I'm just going to accept 2 or 3 people who actually build the product." In fact, in the YC application, it says, "Who wrote the code for this? Who's writing the code?" just so you can make sure that you're actually hiring coders.

What are you seeing on the field in terms of vibe coding? Because people are now—

Great question. English is the new programming language. Only 2% or 3% of the country knows how to code, probably half that code well enough to do a startup. So here we are. Could we be on the precipice of 10 times as many startups, 100 times as many startups?

Absolutely. That's the dream. YC was started 20 years ago based on PG's insight that it's getting easier to start a startup. It used to be that you had to have a big mountain of money, hire a big team, et cetera. His realization was that you can start a startup with just a couple of people—basically, a few kids living off ramen. And that's proven to be true.

Our belief is that with AI, it goes that much further, because the universe of people who are able to create apps using something like Replit is enormous. My most optimistic vision of what we're doing with all the AI is essentially putting all of these tools of wealth creation in as many hands as possible.

Do you think that English is—I think it's Andrej Karpathy who said this, right?—the ultimate destination language that everybody will use to code? Or do you think it gets abstracted even further beyond that, where you sort of think things and they just appear?

I think it might be a little while until we can just think them. But clearly, that's the direction, right? You have a dialogue with the AI, and so you describe, "Okay, not quite like that, more like this."

The direction is essentially just that it becomes easier and easier for us to realize our visions, and for everyone to realize our visions—not just people who are traditionally able to code.

Well, let me ask you this question. That clearly grows the funnel, right? So now we have 100 million, 500 million, 1 billion people, 2 billion people—whoever can speak English can now code. How do you think about that as one of the best computer scientists that America has ever created?

How do I think about all those people having the ability?

Yeah.

I think it's great. Our philosophy is that I don't want to see all of the power concentrated in a small number of large organizations. I think that's bad for everyone. It's bad for freedom.

And so what we want is to give that power to as many people as possible so that everyone can create apps. It might just be something for their own local community. Not every one of those apps is going to be the next Google, obviously, but the more people can create wealth in their own community and in their own lives, the more we spread prosperity everywhere.

Are you seeing in the applications you get to Y Combinator, or that you've heard of, more physical AI, robotics, automation—those sorts of tools? As this becomes easier, it actually leads to the leap: hey, maybe I could do this as a robot, and I get a robot to do a particular thing, and that creates an opportunity for a new business.

Has that become a big growth curve right now—physical AI and robotics?

Absolutely. The number of robot arms at the most recent Demo Day was striking. I think everyone is starting to work on that. Again, as the things that used to be difficult get easier, we just start doing more difficult things.

All technology curves—

Yeah, exactly. And I think that's going to open up whole new realms that were previously impossible or impractical.

So does that create new industries? I think that's a key point. What I think is most misunderstood about AI is that it's not about the displacement of doing old things, but about activating new things that are complex and historically not tractable, but now they're tractable.

Absolutely.

Right. Exactly. So, if you think about just the fundamentals of wealth creation, the inputs are essentially energy and intelligence, and we're about to unleash an abundance of intelligence where the total global intelligence is going to 10x, right? And so, that will enable us to 10x our total wealth. That's going to come in a lot of different forms.

As we start to have AI science labs, for example, where AI can actually start running its own experiments and producing its own data, I think our understanding of biology is going to be incredible. In 20 years, we'll be able to know how a drug affects the body without ever actually testing it. My prediction is that our AI models will be more predictive than today's clinical trials.

You know what's interesting hearing you talk about this, Paul, is the power of great conversations. There was a trope over the last couple of years when somebody lost their job in journalism: “Learn to code, learn to code.” And now you think about it, there are multiple types of intelligence. Startups were limited or gatekept in some ways by mathematical intelligence—the ability to write code.

Opening that up to people who are high-intelligence, high-design, or high emotional intelligence could lead to many more beautiful, interesting products that maybe people focused just on mathematical intelligence would never get to.

Absolutely. This is an abundance that I think people are maybe not even realizing yet. A whole group of journalists and writers who are being displaced, or Uber drivers, or people working in factories—if they can embrace this technology, and we saw it with no-code. Remember the no-code kind of ghetto that emerged for a couple of years? “Startups are going to be no-code.” It was kind of like the false start, but you did see a bunch of new entrants applying for Y Combinator or other things.

This could really be accretive to humanity.

Yeah, absolutely. And it reaches people who are perhaps otherwise left behind, right? It shouldn't be just people in Silicon Valley who can create apps. There's a whole country full of people who have ideas. The same thing goes not just for apps but for media.

I think a lot about, again, when we look at where generative video models are going, it's pretty amazing, right? In a couple of years, that means a kid in middle America or the country who has a vision for their own Disney movie can actually just create the Disney movie. You don't need the $100 million budget. That's going to give a lot of voices that are currently not represented in media because they don't have access to the capital or Hollywood.

The elite version of this would be, “Oh my God, we're losing this job creating at Netflix,” but you're creating a million other jobs for people to create their own superhero that represents them, represents their country, and represents their sensibility.

Exactly. Let me ask you a question as a technologist for a second. When you see the landscape of these foundational models and how good they're getting, is your belief that the number of those will grow, or do you think that they'll consolidate and there'll just be fewer but better? How do you see all of this investment that's happening now—

Yeah.

—play out? And feel free to name companies while you're doing your analysis—which ones will go away?

Yeah. No, I expect that it'll probably stay relatively stable, honestly, because the cost of building these foundation models is astronomical, right? We just saw xAI is raising another $20 million, something like that. The capital requirements are going to limit how many there are.

But I certainly hope that it doesn't consolidate down to just 1 or 2, because, again, I think part of what's important for preserving freedom is just that we have many options. A lot of people don't know we started OpenAI at Y Combinator 10 years ago, in 2015. We saw that AI was on the rise. We saw that this was happening, but at the time we were concerned that it was essentially all locked up inside of Google. Facebook had also had some missteps with Llama.

That would be bad, arguably, for the world, but certainly for our companies, right? We have thousands of companies. If our companies don't have access to that next wave of technology, we're going to be out of business. OpenAI was kind of a moonshot project that we were actually going to take out so it wasn't just locked up inside of—

How did you feel when they made it ClosedAI?

There was never specifically a promise for it to be open source.

Sure it was.

But, again, I think what's most important is that we actually just have a lot of choice, right? I certainly support open source because I think open source is the thing that—

You think open source wins?

I think we'll have both. It seems like the balance is that there are reasons to have both, but the importance of having open source as an option forces all of the closed-source vendors to be honest, right? If they start censoring the models or disabling too many abilities, then people will all switch to the open source.

Well, you worked at Google, you worked at Facebook.

Oh, this was my question. Google has done an incredible job with their ensemble of Gemini apps—I mean, Gemini models.

Or Google?

Well, I'm actually more curious about Facebook. Are they making the right bet with respect to just the talent war that's been created, or is there a different technological approach? The one thing that we talked about before was this concept of the bitter lesson, which is always that compute overpowers humans. How do you think about that, or what would you do if you were running that business today?

I mean, I think he's doing what needs to be done, right? Facebook has clearly fallen behind, and that's a real threat, right? Facebook actually competes with AI. People are switching from Instagram to ChatGPT. My kids are not on social media; they're talking to the AI.

It's fundamentally cannibalistic is what you're saying.

Yes. Yes.

That's an interesting concept. It's a finite amount of time, and we forget about the categories we put on them. I mean, the compound question of asking a great agent is incredible. You know, the way that you can speak. And now, with Grok having the avatar, they're kind of leaning into this concept of personality. We as old people and Gen Xers might be totally missing the script.

Sure. Well, actually, Character.AI is an example that made that bet. A friend from Google who basically invented Transformers got frustrated that he couldn't launch anything at Google, so he started Character.AI. That was the entire thing: making characters that people want to talk to.

Well, thank you for being here.

To be continued. We have to have you on the pod. New fund—that's amazing.

Yeah. Congratulations on the new fund.

Thank you.

Yeah.

Thank you, Paul. Appreciate it.

Thanks for being here today. I'm not sure you've been following the panels, but there's been a lot of conversation around AI, particularly around job displacement. You're the 28th administrator of the SBA. I think more than half of the American workforce is employed by small businesses or is made up of small-business owners.

You and I had a conversation a week or so ago about what you're seeing on the ground with small businesses in an AI workplace setting. The conversation is always: Are they going to get outcompeted? Are they going to get displaced? What's going to happen to American jobs and to small businesses? What are you seeing on the ground, and how does the SBA associate with the transition underway?

First of all, great to be here. Look, small business is big business in America, but small business is big business for AI, and I have been walking hundreds of factory floors for the last 6 months. Most manufacturers in America are small businesses, and without AI, we would not be winning back these industries.

I'll just tell you a case in point. I actually brought a slide to show you workforce development in action. Modern workforce—we call it the new-collar boom. I don't know if they can put it up, but it's a factory in Seymour, Indiana. It's a bike factory. We had lost the bike industry over the last 30 years: thousands of jobs, 98% imports.

We're now, for the first time in this country, building bikes in America because of AI and advanced manufacturing techniques.

Imagine we replicate this industry after industry, and these are small businesses. This is a 60-person factory in Seymour, Indiana, where there are no jobs.

So AI is a job-creation machine for reshoring, onshoring, and advanced manufacturing. In manufacturing, you're seeing a big influence and potential for redefining the industry. What about in the services businesses? What do you see there?

Across the board, we have 7 to 12 million jobs open in America. Most of them are open at small businesses. The number 1 concern of small businesses is a skilled workforce.

That's because President Trump solved inflation, regulation, and taxes. Now they're saying, “Okay, we're booming. We've got $15 trillion of investment coming in. A lot of that's going to trickle down to small businesses. We need the skilled workforce.” President Trump is ensuring that we have that skilled workforce through some of his workforce initiatives, but small businesses are going to be driving the AI boom from the bottom up.

What is needed for workforce training and transition?

Technology is going to be a big part of it. When you go back to 1940, our workforce size was 56 million, and people say, “Well, as technology advances, our workforce gets competed away.” Today, our workforce is 170 million, and compute power has been astronomical.

Essentially, 85% of the jobs that exist today have been driven by advances in technology, and only 40% of the jobs that we had back in 1940 still exist today. We are relying on innovation as a job-creation engine. It's just that people have a fear of the unknown, and they're saying, “I can't envision what it is.”

I can't envision what my life would have been like when I started a small business if I could have had Figma or Canva instead of PowerPoint. Oh my gosh. We're going to create millions of solopreneurs who are going to have massive software companies or manufacturing companies, thanks to AI.

Is there something the government can do through the SBA, and what is the role of the SBA? I know one of the big focuses of this administration was to make government smaller. Is that a goal you have—to make government smaller—and then maybe give the ability to give loans to the states? What is the role of the government in getting 1- and 2-person companies up and running, if anything?

The mission of the SBA is to grow the economy and to support small businesses, and that's what we're doing. For the last 4 years, it had not been doing that. In fact, with regard to AI, the Biden administration banned the use of SBA-backed loans to purchase AI technology.

I had the rules rewritten. Now small-business entrepreneurs, solopreneurs, and 500-person factories can use the proceeds of their loans toward AI implementation and advanced manufacturing. Our goal is to get out of the way.

Educate us on the loans, because we're in venture capital, where we have an incredible ecosystem of angel investors doing this. How do SBA loans work? Who are they for? How much do the American taxpayers put into this, and what's the result?

I'm glad you asked. The SBA does not do direct lending. We work through a network of thousands of banks in this country that offer SBA government-backed loans.

We also operate the Small Business Investment Company, or SBIC, program, which has been responsible for backing many massive startups. SBIC money was in Tesla, for example. We have an equity piece as well as the SBA loans, but those loans have to be repaid over 30 years. They simply give small businesses that banks wouldn't normally lend to a government guarantee that gives them confidence.

We do about 2,000 Main Street loans every single week. So far this year, we're on pace for a record year because we've made the SBA right-sized, which means we've taken it back to its pre-pandemic size. It had doubled during the pandemic. Ninety percent of the employees were working from home and not focused on small businesses. We took it back down, and the spending had doubled, so we took the spending down. We took the headcount back to pre-pandemic levels, and now we have record levels.

You have people showing up at the office.

Oh, yeah. We're back every day.

Wow. So the American taxpayers are paying people for a job, and they're doing it in an office.

Not only that, outside of Washington, we sent them out to the field.

Do you think that at some point you will look at either adding new types of SBA-backed loans or changing some of the conditions to, as you said, further incentivize investment in AI?

Yes, absolutely. We are looking right now at critical industries like medical, metals, minerals, and medical devices for reshoring and onshoring. We have a massive effort at the SBA: we're leading the Make Onshoring Great Again portal, which is on the SBA website. It's a resource for 1 million onshore manufacturers.

We're leading the Made in America charge, so we're focusing on smart manufacturing and looking at loan types. We're trying to double the size of SBA loans so that, for buying advanced technology equipment, CNC machines, and training, there are many more resources available.

How do you think about energy? Yeah. On top of that,

I was just talking to Secretary Burgum and Secretary Wright last night at the White House, and we were talking about the convergence of small business with the physical and the digital. Energy is going to be a big part of that for small businesses because the innovation is going to be coming from smaller businesses.

In manufacturing, you can be a small business and have 1,500 employees, but frankly, I'm seeing a lot of energy companies and others with 300 people. So small businesses are going to drive it. If you stipulate that there are 34 million small businesses in America and 20,000 large companies, this is a small-business-driven energy and AI boom.

Your vision is something that some of the leading entrepreneurs in Silicon Valley have been pushing for as well: this idea that there is an entire boom that will happen of solopreneurs—the 2- and 3-person companies that are vibrant, successful, profitable, and growing. What they just need is a little bit of help at the edges, potentially on paying for some compute resources or whatever, and then they're off to the races.

That's certainly backed up by the data we have at the SBA. Sixty percent of the $21 billion that we've lent this year has gone to companies with 1 to 5 employees. That's where the growth is coming. Certainly, we know that they're going to scale from there.

We're seeing all the trends say that putting more technology into the hands of small businesses is growing the economy, and small businesses are still driving the jobs boom in America. There were 720,000 jobs created this year, led by small businesses.

Keith, I'm curious. You're a free-markets guy. What are your thoughts on the government's role in juicing up the onshoring, specifically in categories where maybe China has dominated for a couple of decades?

Keith

As Kelly pointed out, the government is actually not extending the loans. The community banks in America are extending the loans, so it really isn't a deviation from free-market principles.

If you think about it, AI is really rocket fuel to turbocharge small businesses and entrepreneurs in at least 3 dimensions. First, access to information. Typically, if you're starting a business, you have to compete with very large incumbents that have expertise in market research, marketing, legal, accounting, and more. Now, at the tap of your fingers or your voice, you have the same expertise that all these large companies have. So you've leveled the playing field.

Secondly, you have access to products like building an app. Everybody can compete with a large company. Anybody can code an app. So if you're an HVAC repair person, you have an app that's on par with a Shopify store or better. That allows you to compete. We're going to see more proliferation there.

Third, you can save money. You used to have to have a G&A team—you'd have accountants, bookkeepers, and HR. AI can do all that, maybe even do it better than humans, but certainly at zero cost. The economics of running a small business are going to be much better.

The risk of running a small business, of starting a small business, is going to go down, so we're going to see an increase. Finally, you can save money through things like Ramp. You can use AI to audit your expenses and not waste 5% to 15%, which will make you more successful.

All these trends are going to combine, and we're going to see an explosion of successful small businesses in this administration.

Does that mean that there are just more competitive forces in the marketplace? Big companies are going to have more competitors, and it ultimately drives net productivity gains long term?

Hopefully, net productivity gains. Insofar as there's some substitution, I suspect you wind up with a barbell. The largest players, the NVIDIAs of the world, do benefit the more people that run compute, et cetera.

But then I think that the smaller businesses actually eat at mid-market companies because they can compete now, and they've been at an economic disadvantage for decades.

We're going to be in industries that we couldn't have even imagined we would be in. When people say, “Why do we need to make bikes in America?”—because it creates 60 great-paying jobs in a tiny town in Indiana, with people who want to do it.

That's right. That's right. PPE and pharmaceuticals during COVID—we should be making that here.

We can do that with smart manufacturing with 100 people in the factory.

You must give the criteria or some guidelines to the banks on how to pick. I'm assuming you take diversity, inclusion, gender, and all these important factors into account, or do you do it based on merit?

I was just trying to trigger the two of you. They send that to your fund.

Yes. I don't do any DEI, but jokingly. But what's the criteria when somebody comes and says, "I want to raise $100,000," and goes to their local bank? How do they get picked?

We have strict underwriting guidelines, and we've stripped out the DEI that the last administration had put in. They had a Green Lender Initiative to preference where money went under the Green New Deal. We've gone back to saying, if you qualify for these loans, have at it. We're not going to pick winners and losers. We want everyone to compete on a level playing field and have access to that capital.

But what had happened under the last administration was that they'd lowered the underwriting guardrails. As a result, the loan losses on the portfolio went way up—$400 million. We've reversed that and strengthened the underwriting standards to make sure that the money goes to small businesses that are building these factories to onshore drones, pharmaceuticals, defense, and aerospace.

What are the target performance ratios in the loan portfolios?

Oh, my gosh. Our loss ratio should be 3% or less, and it is. The SBA is one of the largest disaster lenders in the country and a disaster-recovery lender, and they do very well—very low, very low. In fact, there's a secondary market for SBA loans because they perform well due to the strict underwriting standards.

Zero subsidies—sorry. It operates at no cost to taxpayers when we enforce prudent underwriting standards, which we're getting back to.

One of the things that helps burnish entrepreneurship is that imitation is the sincerest form of flattery. You must have so many successes, but they're not always well marketed or known, which would pull other people to say, "If they could do it, I could do it." How do you think about that in a world of social media and all of this?

You've picked up on one of my key problems. I run an agency that starts with the word "small." Small does not mean insignificant. In fact, small business is significant, and President Trump and I talk about that all the time. He loves small business. He knows innovation starts there. Manufacturing is small business.

So, we're working on a massive resetting of what the SBA does, but more importantly, what small business means to America. I think people are waking up that Main Street is going mainstream, and we have to continue to push the understanding that if we don't protect our small businesses, our innovation pipeline and our job-creation engine are going to shut down.

How do you interface with state agencies, state senators, and state governors who have 50 different views of the world, but you're responsible for at least supporting the underpinning of the business people who are there? How does that tension play out?

It's really important. In fact, we've started an initiative where I'm meeting with governors across the country and their economic development departments, essentially, because they know best what they need in their state. If we can push more of this out of Washington and say this needs to return to the states, they need to know the SBA is a resource for recruiting companies into their state to create jobs in manufacturing, like in my home state of Georgia, which has done that.

We're going to continue to partner at the state and local levels and across the administration. Having David Sacks in this administration as an ambassador for AI and crypto has been huge because it gives us a way to work across the administration, and then we can focus with the governors at the state level.

Can I ask one question on that? Your comment was really striking: You guys have strong underwriting performance in the loan portfolio. There are many other insurance programs across the federal government that do not have good underwriting standards, run terrible loss ratios, and are highly inefficient for the taxpayer. Then they cause all of these market inefficiencies as a result.

I won't start to name them, but you know who they are. Given your background in financial services and fintech, and your experience here, is there an opportunity—do you get drawn in—and is there an opportunity to go in and try to address some of these other very, very, very large insurance programs and underwriting programs that the federal government operates?

I think there is, because we've recruited to the SBA really an elite group of financial-services leaders who understand this. I served in the Senate previously, in the U.S. Senate. I was the only CFA to have ever served in Congress, and I found out when I went to Washington—

Congress ever?

Ever. They don't like people with financial-services experience in Washington because we know how to read a P&L. But we're bringing that discipline. We're happy to share it. It's very open source. Please do. So, like you say, we've open-sourced it, and the fans have just gone crazy.

Kelly, we have a word for small businesses in our community. It's called startups. Maybe it's time to rebrand the SBA.

I'm completely open to it. That's right. I was going to call it Main Street Manufacturing, but I like startups a lot, too. I love that.

By the way, the point about China we made earlier is that they have 3 million factories, but these aren't massive 100- or 400-acre facilities. These are very often small warehouses that were turned into small manufacturing facilities, and we could recreate that in America across all of these great states where people are looking for economic expansion.

Over and over. Well, Kelly, actually, many of those opportunities exist, but people don't know that they can find local sourcing.

And you can use AI to go across the entire country and find local manufacturers. There's almost always a choice in the United States. It's just that people don't know where to find them, how to negotiate with them, or even how to get in touch with them. That's a solved problem now through AI.

Yeah, yeah, yeah. Do you think there's a place where the SBA, maybe in partnership with the White House, says, "Here are these industries that are just a little bit more important, or kinds of companies that are just a little bit more important for a bunch of strategic reasons," where maybe you relax the underwriting criteria or just try to get a lot more people on the field, chips on the table? How do you think about that?

First of all, I'm a taxpayer champion because, as a small-business person, I know that small businesses are taxpayers, too. We can't put some small businesses on the hook for other small businesses. We've got to have an efficient market that discovers the right funding mechanism.

We're looking at making sure that we have the right underwriting standards for critical industries. We're working on some things with the Department of Defense right now. We have our SBIC program, where we're experimenting with some different equity structures, so there's more to come on that. I think financial engineering is important, but first and foremost, we have to not put taxpayers on the hook for it.

I think that's a really interesting point. The equity structures—if you look at Solyndra, Tesla, and that cohort, Tesla paid back its loan with interest early. If the government had gotten just 10% of that in equity, that would have paid for 100 Solyndras and mistakes. So, some equity component or warrants could change the SBA into having—and the American taxpayer, by proxy—some upside in these investments.

You like taxpayers having equity?

It's tricky. It's more complicated than that. You have adverse-selection issues, and it's not a one-size-fits-all situation. But having flexibility for certain industries to have a different corporate structure or different investment structure is Pareto optimal.

You don't do that at all today, Kelly, at the SBA?

Not today. It's very plain vanilla. But we're continuing to have the conversations about how to be creative, particularly around defense-critical technologies. There's a lot to do there, and we need to do it very quickly. There are some great success stories that we can replicate, and they may not even require massive reengineering.

In the last few minutes, Kelly, can you give us a very quick contrast between your life as a senator and your life as head of the SBA?

I'd much rather be an executive than a politician. I was humbled and honored to serve in the Senate. As a kid who grew up on a farm and was the first in my family to graduate from college, it was amazing.

But being able to run this agency, which at 7,000 people is considered small, is amazing. I'm really an entrepreneur and a businesswoman at heart, so I'm approaching this as a businesswoman in service to taxpayers and the government. I'm incredibly blessed to be able to do it.

So, I love it. Thank you for doing it. Please join us in thanking Kelly.

Thank you. Great to be with you. Thank you.