Winning the AI Race Part 3: Jensen Huang, Lisa Su, James Litinsky, Chase Lochmiller

This is one of the most amazing entrepreneurs you're going to meet: James Litinsky, founder and CEO of MP Materials.

Thanks. Good to be here.

How are you? So, let me set this up. James was a hedge fund guy running a pretty successful hedge fund, and he ended up investing in something called Molycorp, which went out of business.

Yep.

Yeah. And you did this incredible thing: You said, “You know what? Screw this.” You essentially shuttered the fund, took over the company, and fast-forward many years later, you are the largest and only supplier and refiner of rare-earth materials—and maker of magnets—inside the United States.

We're 100% of the American industry.

100% of the American industry. You just did 2 really incredible things in the last couple of weeks. One was announcing an enormous public-private partnership with the Department of Defense—$400 million, et cetera. And the second was announcing a really big deal with Apple.

Yes.

Okay. Take us a huge step back. Talk to us about why rare earths matter. Tell us about the supply chain for AI. Tell us why you're doing this.

Rare-earth magnets are really the feedstock for physical AI. Robots, drones, everything we're talking about today—the biggest industry in the world to come—essentially, electrified motion requires rare-earth magnets.

The predecessor went bankrupt. There was a feeling when I took over this site with my co-founder. This goes back to 2015.

Where is the site?

It's in Mountain Pass, California. If you take a 45-minute drive from the Las Vegas Strip, just over the border in California, you can see the site from the road. It's really the best rare-earth ore body in the world.

The thing about rare earths is that when you mine them, you also have to refine them. It's really expensive and difficult to refine them. It's a specialty chemical process. Think of it as a multibillion-dollar refinery that you need just to separate them. Once you separate them, you need to turn them into metal and then a magnet.

There are multiple layers to this supply chain. Of course, you could have all the rare earths in the world, but if you don't make the magnets, you're sending it to China. Or you could have all of the magnetic capability in the world, but if you don't have the rare earths, you're relying on China.

Our vision from day 1 goes back to when we originally bought these assets out of bankruptcy. Officially, it was a 2-year battle; we took it out in 2017. There was a perception that we just couldn't compete against China. What we discovered is that we could. It's a world-class site, but we had to reorganize the process flow, and then we had to make investments to move downstream.

Over the last 8 years, we invested about $1 billion. We took the company public in 2020 and built out the refining capability. About 4 years ago, we announced that we were going to build a magnetics factory in Texas. We built that factory, and we have GM as a foundational customer. We're now producing automotive-grade magnets to GM specifications, and we'll be ramping up sales to GM at the end of this year.

It's been a busy few months for us. We announced a pretty transformative public-private partnership with the Department of Defense. DoD is becoming our largest economic investor, and they're also going to provide a price floor for our commodity so that Chinese mercantilism—we can get into that—won't take the price of the commodity below the cost of production.

As a result of the DoD investment, we're going to accelerate the buildout of the magnetic supply chain. We're expanding our facility in Texas for Apple, and then we're going to build a 10x facility to 10x our capacity, with DoD as our 100% offtake partner, customer, and business partner. We'll be splitting profits 50/50 with DoD.

To just translate this, it's not a handout from the government. They didn't gift you $400 million. They invested in your company. They have warrants. They have equity.

Yeah. They invested. They're both an owner and an upside participant in our commodity to the extent that prices take off. They're also a 100% offtake customer. We have a guaranteed level of profits to warrant building out this facility, but above a certain threshold, they're a 50/50 economic participant.

So there's really you, the taxpayer. So this is a true win-win. Obviously, it's great for MP shareholders. It's also great from a national security and commercial-national-security standpoint because we're going to have enough magnets to provide real certainty in the supply chain for the physical-AI revolution and other industries.

It wouldn't surprise me if, 5 years from now, we'll do this conference and you'll say to me, “James, I remember that deal that was the first of its kind that you did with DoD.” The government made money on it—the taxpayer made money on doing this. I'll say, “Yeah, I actually think that's going to be the outcome.”

There's an element of mutually assured economic destruction. If the Chinese believe that America has national champions, then there's no point in subsidizing the rest of the world. I think you can start to see prices normalize for some of these things and free up our ability to invest and expand.

Why go to the government for this investment as opposed to the private markets?

Because it's that issue. This is one of those situations where you have to go back to World War II or the railroad boom, where you really need government and credit. This administration did something totally unique that—

Why do you need the government?

Mercantilism. Straight-up mercantilism. The Chinese will sell magnets for below the cost of raw materials. Every time there's somebody who makes progress, they can put them out of business overnight.

It's difficult to want to make the investment. Frankly, with the Department of Defense, given the scale they wanted us to build and the time frame they wanted us to build on, there was no way we were going to make that commitment. We're fiduciaries; we have shareholders. There's no way we're going to make that commitment without certainty that we would not be destroyed by mercantilism and that we would have a customer for the magnets.

How big of an industry is physical AI? We see the robots, and we're told the robots are coming. We're told there are going to be billions of them. Are they actually being deployed at the scale and at the pace we've been told?

I think that's a question for much smarter guests. I will say that one of the big drivers of our deal was, as we've seen in Ukraine and the Middle East, the future of warfare is physical AI: robots and drones.

Irrespective of the scale that robotics ultimately reaches—and certainly the commercial business will be bigger than the defense needs—from a defense standpoint, this is a really important supply chain that we must have. We can't be funding cutting-edge drone and robotics companies and then say, “Okay, but we're going to buy those magnets from China.” That makes no sense.

Do we have the talent capacity, or do we have a talent shortage? Secretary Bergam gave me a statistic that was pretty shocking: We only graduate 200 people a year in the United States in mining, which is orders of magnitude different from China. What do we need to do to be competitive and build the industry here?

It's a great question. I think about this question a lot because—

What's that? Oh, my God. Sorry.

No, it's all good.

I'm a huge fan of the pod, and I embarrass myself all the time.

You know, I'm a fan of the pod since day 1, and I totally embarrass myself.

That's only one correction. I'm messing with you. Was this intentional?

Huge fan of the pod.

Yeah, huge fan of the pod. Who are you again? I'll take a selfie later.

And I'm not the AI guy. Go ahead.

We have 850 employees today at MP. We're going to hire a couple thousand more people easily when we include what we're building out for Apple, coupled with what we're going to build with DoD—not to mention the construction jobs.

This is a key existential question for all of us as we build out. Where are we going to get the talent? What we've found at Mountain Pass is that we hire electricians, maintenance workers, and operators, and we get people in and train them. Then, obviously, you give people a career. We've been training a lot of people, and it's a little bit more painstaking, but there's absolutely talent out there. People are hungry to do it.

Why do you think it's been so hard to establish that idea? You find it straightforward to find good, hardworking people to get into these jobs, but the perception is always that these jobs are not desirable. They really are desirable to many people.

Yeah, absolutely. Our median wage is now pushing $100,000 a year. Relative to some of the opportunity set, these are great jobs.

And what are the salaries?

What’s that?

What’s the starting salary? Just curious.

So, it really depends on the job function. I think the easiest way to think about it is that you can certainly make close to $100,000 a year as an operator with us because, by the way, everybody’s an owner. We have an owner-operator culture. Everyone got stock when we went public in 2020.

But somebody coming out of high school can make $40,000, $50,000, $60,000 or more.

Yeah, or it depends. We can’t find enough electricians, and we can’t find enough maintenance workers. A maintenance worker or an electrician can make 6 figures today.

You said earlier that you suspect 5 years from now we’re going to look back and see that this deal with the DoD was a blueprint.

Yeah.

Give us other areas of either physical AI, software AI or other markets where you think these public-private partnerships are really necessary to establish U.S. supremacy.

There are some major categories. Obviously, we’ve all heard about shipbuilding and advanced pharmaceutical ingredients. I think those are important ones. Then there are a number of niche areas, like industrial diamonds, that are important for quantum computing—some of these things that you never would have thought of, where there might not be a market large enough to need 5 players, but a good public-private partnership can solve that problem. Then there are some other verticals in critical minerals.

Was it straightforward for you to find the right person within the Trump administration who said, “Of course, this is obvious. Let’s sit down and hash this out”?

Yeah. I think that’s because our particular deal was led by the DoD, and I have to say that the Pentagon leadership is extraordinary. This was a mandate directly from the president to solve this problem. Again, they deserve a lot of credit for being bold here. To be clear, because this story is not out there, I’ve never worked so hard in my life. This was a true, aggressive, private-equity-style investment and negotiation.

The transaction documents are public. You can look at that.

So that’s you saying that they’re tough.

Yeah, they are. This was as tough as it gets—tougher than any blue-chip private-equity or distressed-lender-type negotiation. That’s what this was. The key thing was they were going to hold our feet to the fire to execute on an aggressive timeline. They were going to hold our feet to the fire on the cost, so we’re exposed if we get the costs wrong. We’re making this investment.

The key piece of this, which I think is a good model for all of us and will actually be really effective, is that the goal—I don’t speak for them, ask them—but I think their goal was: We’re going to take the things off the table that you can’t control—mercantilism, certain customer issues. We’re going to be held to account for the things that we can control: our ability to execute, our ability to execute on a good timeline and our ability to control costs.

When we think about these things historically, the government is sort of investing in a sector and, quote, picking a winner. Usually, there’s money given to someone, and it’s sort of public risk, private upside, right? This is not that. This is private risk, public risk, public upside, private upside. It’s a true shared win-win-win. Again, hold me to these words: I hope I’m right on this, but to the credit of the Trump administration, I think they will make money on this and solve the national-security problem.

All right, we appreciate you coming, James.

Yeah. Oh, thanks.

Thanks so much.

Thanks, brother.

Yeah, it’s great.

All right. Take care, James.

Thanks, Jacob. I appreciate it. Yeah. Okay.

Hi, Lisa.

Lisa, it’s a pleasure.

Hi. Nice to meet you.

Hi. Well, thanks so much for being here with us today. We don’t have a lot of time, so we want to get into it. In April, it was announced that you achieved your first silicon output at the TSMC facility in Arizona on that 2-nanometer line. This administration and the private sector have talked a lot about onshoring semiconductor manufacturing. We’d love your thoughts on the on-the-ground experience in Arizona. How’s it going? What’s not going well? What does America need to do to get this right?

Well, absolutely. First of all, it’s a pleasure to be here. I love the theme. I think we’re all super excited about winning the U.S. AI race. I thought, if we’re going to talk about chips, David, I should actually bring one.

Oh, awesome.

That’s okay. A little bit of show-and-tell. So, this is our latest-generation AI chip. It’s our MI355 chip. It has 185 billion transistors and takes about 9 months to build. There’s a lot of technology on it.

Nanometer?

This is 3-nanometer and 6-nanometer, so there are lots of different technologies.

I’ll be putting this on eBay later.

I’m going to take it with me when I leave. How is that?

Thank you.

But look, to answer your question, these AI chips are extremely complex. They have so much technology on them. We’re super excited about the progress in U.S. manufacturing. I would say 12 months ago, people weren’t sure that we could do leading-edge manufacturing in the United States. We’ve been very early in Arizona with TSMC, and we did get our first chips out. They’re actually 4-nanometer chips, but what we see from it is that where there’s a will, there’s a way.

I think all of the conversation about onshoring manufacturing has been super good for the semiconductor industry and, frankly, for all of us in semiconductors. We’re in such an interesting place because chips are so essential to ensuring that we’re able to win the AI race. We want to make sure that there’s a lot of geographic diversity and capability there.

What about cost in Arizona? It’s unrealistic to think the United States could compete on cost. Am I correct?

We’re going to pay a little bit more.

Give us the ballpark. 50% more, 20%?

Not 50% more. It’s going to be more than 5%, but let’s call it less than 20%.

So, low—

Low double digits. Let’s say low double digits.

And how does that impact the business, if at all, in terms of competition globally?

Well, I think the important thing is, just think about it: Everybody wants a GPU, right? If you look across the industry, you really say that the people who are going to win in AI want to have as much compute in their foundation as possible, and they want assurance of supply. We want to be able to supply this no matter what happens.

If you put that in context, the fact that you’re not going for the lowest cost every minute of the day is okay. It’s okay. Obviously, not everything needs to be in the most advanced technologies, and so we have a very geographically diverse supply chain. Taiwan continues to be important in that view, but the focus from this administration on getting onshore manufacturing in a big way, not in a small way, I think is very good for the country.

How much time do we have? If there was a disruption, for whatever reason—we can come up with hypotheticals—in Taiwan, and we were unable to get chips from those factories, what would that look like globally?

You have to look across the supply chain, but from a structure standpoint, we all want to keep reserves for those times. But it’s months, not years.

Lisa, there were 2 really interesting posts over the last couple of days. One was from Elon, where he said that in 5 years he projected 50 million H100 equivalents just for xAI. The second was from Sam Altman: They signed a deal for a 4-gigawatt data center and $30 billion a year with Oracle. That just portends an enormous amount of chips and power that are necessary. If you forecast that, how do we actually meet all of it? What needs to happen that’s not happening today inside the United States to actually do that?

Yeah, it’s a great point.

I mean, that’s what we’re seeing. We’re seeing this incredibly large demand for AI, and it’s coming from Sam and Elon, who are certainly a couple of the leaders. There’s a lot of demand elsewhere, too. If you think about it, nations want their own AI, so there’s very high demand. We’re imagining that just the accelerator market—the chips for these AI large-computing systems—will be over $500 billion in a couple of years. So, very high growth.

And when you say, “What do we need to do?” it’s the entire ecosystem that needs to scale up. Certainly, what we’re doing in chip design is trying to get chips out as fast as possible. But we’re also scaling up the entire manufacturing ecosystem. And as I said, I think the U.S. is going to be a huge piece of it. So it’s not just about the silicon. There are all of the various other pieces of the ecosystem that have to come to the U.S. And I think—look, I think today’s AI Action Plan is actually a really excellent blueprint.

And how do you see the market evolving over the next 5 or 6 years? Is there a standard set of chips for training, a standard set for inference, or do you just see an explosion—a Cambrian explosion—of different ASICs, different designs, different use cases?

Yeah, I like that question because I am a believer that there will be diversity of chips. The reason is there are so many use cases, right? If you think about use cases, whether you’re talking about science, manufacturing, design, back end, or, frankly, personal AI, I think we’re going to see AI in everything that we do—certainly in your phones and in your PCs.

So you have all these pieces. You’re going to have different types of chips that do that. Certainly, for the largest systems, we tend to believe that you need the most compute you can get, and so GPUs are there, but lots of ASICs are in the process, and we’ll see a variety of different chips.

You opened up a really interesting line of questioning there. When mainframes were so expensive, and then eventually we wound up having PCs on our desktops, you alluded to AI being run locally.

Yes.

When would we have a local computer—a laptop or a desktop computer—that would have the power we’re seeing to run some of these LLM models, in your mind? And do you see that as a specific market to go after?

I definitely see the idea that AI will be at every part of our ecosystem as a real thing. I think that’s one of the advantages. If you think about the power of AI, you want it everywhere, and you want it across all different applications. And I think when you think about PCs today, we’re putting a significant amount of AI in them to run local models. And why would you want that? Well, maybe I don’t want all my personal data all over the place.

On that point, can you make a prediction on when the market for physical AI chips is greater than the market for chips in data centers?

That’s a great question. I’m a big believer in physical AI. I still think it’s—let’s call it—5 years.

You think 5 years? Is that that fast?

It’s at least 5 years.

So you’re saying 5-plus.

5-plus. Yes.

Okay.

But that is ultimately the biggest end market.

Do you think physical AI becomes the biggest end market?

I think it becomes a significant end market. I think you look at chips in data centers and you look at chips at the edge; they’re also significant markets.

When you look at the most cutting-edge techniques today—EUV lithography, all of this stuff to make chips—one of the things that’s observable is we’re only as good as what humans have been able to invent. I often ask the recursive question: What happens when AI is able to invent its own method of manufacturing—different materials, different materials science, different approaches that we may not necessarily understand? Is any of that R&D happening, whether at AMD or in other places? How are we trying to get beyond the physical limits of electrons tunneling across a junction?

I think this idea that AI can be extremely smart and extremely capable—when we think about how AI can design the future chips, it will design pieces of it, but there’s still a creativity in bringing it all together that I think humans are still absolutely at the center of. So I don’t necessarily see AI designing our next-generation GPU, right? But I do see it helping us design the next-generation GPU much faster and more reliably.

You talked about the need to reshore more parts of the ecosystem. Obviously, you guys have world-class chip design, and the fabs are getting reshored, but how do you think about things like lithography? Does that need to be reshored, or does ASML need to start building machines in the United States, or is it okay to have that type of supply-chain risk with an ally?

Look, I think we’re going to—we have to accept the fact that it’s a global supply chain. Even if you were to reshore X number of components, you would still have Y components that are across the world. I think it’s important for us to have our allies together. So that’s a key piece of the conversation: ensuring that we have access to the latest-generation technologies. And that is something that we protect, given our intellectual property.

And going to first principles and asking you the open-ended question: What should be done about American education? I’m going to ask this a lot today. Assume there’s no college, high school, nothing. You arrive in America; the situation is what it is today. What do you do? How do you build an education system to prepare the next generation for the evolving workforce?

Yeah, I’m probably a little bit biased, as maybe some of your guests are today. I’m a big believer in a science and technology background—a STEM background—as being so helpful when we think about the future workforce. And the earlier we can get into the process, I think the better.

Some of the work that’s being done to revitalize the curriculum, I think, is pretty important for the next-generation workforce. And one of the things, when I think about how we win in AI—there are so many aspects of it—is ensuring that America is the best place for AI talent. That’s also a key piece of it. So inspiring people when they’re young to really study science.

Lisa, when you go to bed at night and think about the best-case scenario for this technology and this trajectory we’re on, which is accelerating and you’re enabling, what could the world look like in 10 years? Let’s say it’s pretty obvious we’re hitting artificial general intelligence at this moment. I think we’d all agree we’re starting to see that. But superintelligence can’t be far behind that. I assume you agree with that. Assume we hit that superintelligence: What will the world look like in 10 years in the most optimistic scenario if we do this right?

Well, I think the exciting part about it—and I can say this very sincerely—is this is the most transformational technology in our lifetimes. That’s the way we should think about it.

Orders of magnitude.

Orders of magnitude. And the reason is it’s not just going after one aspect. You can actually take AI and make science better. You can take AI and make medicine better. You could take AI and make manufacturing better. You can take AI and make every aspect of your business better.

And so, in my mind, 10 years from now, we’d like to believe that we are really leveraging it to solve some of the world’s most important problems. I like to say that AMDers get up in the morning and say, “How can I use technology to solve some of the most important challenges in the world?” And AI is really our mechanism for doing that.

I have a business strategy question. If we went back 20 years and wrote the tale of 3 companies—Nvidia, AMD, Intel—and then you fast-forwarded 20 years, 2 have just absolutely thrived and 1 has not. If you had made the bet back then, it would have been very inconclusive whether you would have picked Nvidia and AMD. And if anything, there was an amount of inherent belief that Intel had just figured it all out. Can you just tell us the lessons learned of why you’ve thrived and maybe what you take away from their journey to make sure AMD doesn’t play out that way?

Well, as a CEO, we have to be paranoid every single day, right? So we don’t rely on the past, but I think there are lessons from the past. And I think probably the most important lesson that I can say for technology is: You have to shoot ahead of the puck. You have to be thinking. Your question, James—great question—we think about that all the time: How do we shoot ahead of the puck?

And things change. Technology is a beautiful place because you see big inflection points. 5 years ago, AI was around, but we wouldn’t have been able to gather this audience to talk about AI because people would be like, “Who cares?” But the fact is, you had to invest many, many years ago to be where we are today. And I like to say that you will be able to judge whether we’ve done a good job or not by how we perform 5 years from now.

The decisions we’re making will take 5-plus years to play out. But that’s the key thing in tech: nothing is fast, but hopefully it’s quite lasting in what we do.

What do you think is happening in countries outside the United States? What do you think is happening in chip design and all of these tech capabilities in China and other places right now?

We should believe that it’s super, super competitive. At the end of the day, I think the world has recognized that semiconductors and chips are essential. They’re essential to national economies. They’re essential to national security. So assume that everyone’s investing.

I’d like to believe that we have a great head start because of the innovation pipeline and the great companies that we have here, but we should not be confused: everybody’s investing, and we need to keep up our investments as well. I think that’s why this whole idea that any one company can provide every solution that’s necessary just isn’t the case, right?

I love the idea of open ecosystems, of companies collaborating, of collaboration across the ecosystem—hardware, software, systems—and collaboration across public-private partnerships. That’s what it’s going to take for us to win. We have to be front-facing and realize that the countries that win bring all of the smartest people and the best capabilities together and let them go as fast as they possibly can.

Great. So, thank you for being with us.

It’s been a wonderful afternoon. It’s been great. I appreciate it.

Thank you.

Great. It’s been a pleasure to meet you. Thank you.

I’m Chase Lochmiller, the co-founder and CEO of Crusoe, and I’m here to talk to you about the AI industrial revolution. I’m going to start with a quote from Warren Buffett’s 2020 shareholder letter to investors. He said, “In its brief 232 years of existence, there has been no incubator for unleashing human potential like America. Despite some severe interruptions, our country’s economic progress has been breathtaking. Our unwavering conclusion: Never bet against America.”

Buffett’s words were true then, and as we enter this global race for technological dominance of artificial intelligence, they ring even truer today. American dynamism has always prevailed, and it will continue to do so. In the history of what really made America great, we live in a nation that’s the freest nation in the world, and we’re just as rich in land and resources as we are in human ambition to drive progress.

One of the things that’s fundamentally enabled that progress to happen and that ambition to be unleashed is the leading investments that we’ve made in infrastructure over the course of his lifetime. Warren Buffett got to witness investments in power, transportation, and natural resources to enable people to pursue their dreams and live a better life.

Now, in 2025, we stand at the start of a new era of infrastructure: the infrastructure of intelligence. It’s driving the biggest capital investment in human history. This investment is being led by the hyperscalers, who are investing hundreds of billions of dollars per year to make this happen. These are the companies with the biggest balance sheets in the history of business, and they are quite literally going all in to make this happen.

They’re not the only ones. There are also startups like Crusoe, and there are even nation-states that are following suit. So what’s going on there? What’s the prize that they’re going after?

The opportunity here is that, for the first time in human history, we’ve actually been able to manufacture intelligence. Intelligence is the scarcest economic resource in the history of the economy, and for the first time, we’re actually able to make it. The opportunity here is to unlock access to what has historically been that scarce economic resource.

This is why the data centers of the future are not being referred to as data centers. They’re actually being referred to as AI factories. It’s a factory that takes as inputs data, algorithms, chips, and energy, and outputs intelligence. This is the alchemy of intelligence.

This newly manufactured intelligence will spawn a new chapter of unprecedented productivity and development, and that will serve to improve human quality of life. The IDC estimates that AI will generate $20 trillion in economic impact by 2030. So even if you can earn a small slice of that, those hundreds of billions of dollars of investment will earn an amazing return.

Each dollar invested in business-related AI is expected to generate $4.60. As my friend Jensen would say, “The more you buy, the more you save.” Or, in this case, the more you buy, the more you make. We can grow the pie together and usher in a new era of AI-driven abundance.

When we look at the history of American energy production and consumption, as the United States industrialized, we really ramped up energy generation and consumption. But if you look at this chart, you can see that it’s flatlined over the last 20 years, where we’re generating and consuming about 4,000 terawatt-hours per year.

AI is fundamentally transforming this demand picture, and energy is quickly becoming the bottleneck to growth. Data centers are forecasted to account for 20% of the growth in power demand between now and 2030. Data center power consumption is going to go from 2.5% of U.S. power consumption to 10%.

What this means is that the technology industry that’s willing this infrastructure into existence fundamentally needs to bring its own power to support that growth. That means massive investments, not just in data centers but also in the energy infrastructure to support them. This will require people—lots of people—to build, operate, maintain, and run these large-scale energy investments.

If we look at data centers by the numbers, I think it’s important, as people are throwing around gigawatt-scale data centers, to look at the amount of data center infrastructure that exists today. Northern Virginia is the center of the world for data centers, but at the end of 2014, it was only 4.5 gigawatts. Today, we have companies that are looking at building single 5-gigawatt facilities. If you look at this growth, we’re building more than a Northern Virginia every single year in the forecasted future.

So if there’s one thing that you’re going to take away from this presentation, it’s that we need new infrastructure. We need lots of it, and we need lots of people to build, operate, and maintain it. This is what Crusoe is focused on solving.

Crusoe is in the business of activating energy for intelligence, of building and operating AI factories at scale—from the steel to the silicon, from the electron to the token. If you look at our pipeline, we have about 40 gigawatts of capacity that spans all sorts of energy resources, from new energy technologies like small modular reactors to renewables and natural gas, to power this innovative future.

Revisiting my formula here, I think we left off one critical component, which is the people. AI infrastructure will be the largest job-creation catalyst that we’ve ever seen. I think it’s important to look at what this looks like in practice.

For the last year, Crusoe has been building a large-scale AI factory in Abilene, Texas. Speed is paramount. Again, this event is about winning the AI race. In order to win a race, you really need speed. Crusoe has been focused on using modular components and rapidly scaling investment in construction and infrastructure to support this.

We’ve built a lot of different modular components in factories and brought them to the site. They’re like Lego blocks that fit together to build one of these AI factories at rapid scale and speed.

If you look at what this looks like today, this site will consume over 1.2 gigawatts of power and 400,000 NVIDIA GPUs, all in a single coherent cluster. This will essentially be a gigawatt-scale computer to drive human progress forward. It’s really amazing what you can accomplish in a year. Just 1 year ago, this is what the site looked like, and this is what it looks like today.

What does this mean from a jobs perspective? We have 4,000 people working on-site every day to make this facility happen. It’s a bunch of different trades: electricians, plumbers, and construction workers. It’s required a lot of capital, too. We raised $15 billion to basically build this facility and bring it into existence.

It’s also required manufacturing, and a lot of the critical components have been built off-site in these controlled manufacturing environments. But this isn’t the only one. This isn’t a one-of-a-kind. We’re also building AI infrastructure and AI factories across America.

This site in West Texas is going to be a gigawatt facility behind the meter, with wind, incremental gas, and grid interconnection. We did a partnership with Redwood Materials where we built the largest microgrid in the United States, with 60 megawatt-hours of end-of-life EV batteries and 20 megawatts of solar to power an AI factory.

We have a partnership with GE Vernova and Engine No. 1 for 4.5 gigawatts of new gas-generation capacity to power future AI data centers. And finally, we want to announce a new partnership that we're doing with Tallgrass Energy in Wyoming that will initially power 1.3 gigawatts of total compute load, alongside 2 gigawatts of power generation, and ultimately, we feel like this can scale to 10 gigawatts of power. So we're really thrilled to partner with Tallgrass.

As a vertically integrated AI infrastructure company built here in America, we believe that AI factories will be the ultimate economic engine, creating utility for society and new jobs for the economy. This will usher in a massive new era of AI-driven prosperity for the United States. And I want to leave you with my final quote from Warren Buffett: In this AI race, never bet against America. Thank you.

So, is this stuff real? You guys started off as a Bitcoin miner, and now somehow all the hyperscalers are asking you to build nonstop data centers. Why you guys?

I think, again, it comes back to this being a race, and one of the things that Crusoe has been able to do better than anyone is execute at speed and scale.

I know some of the biggest constraints have been water, energy, and land for this type of stuff. What parts of the country have you guys actually been able to do this in? Have you seen any of the local regulators start to step up to make this stuff easier for you?

We've been building quite a bit in Texas. Abilene, Texas, is this initial facility that's gotten a lot of coverage. We just announced another facility in Texas. Wyoming's been a big area of investment for us. But there's a number of other states that we're evaluating and investing in to build large-scale AI.

Is it only going to be the more rural, red states, or do you think Oregon, Washington, and other states will start to get together and realize they've got cheap hydropower and cheap water and try to get you there?

No. Believe it or not, we're actually looking at something in California.

Wow. California. Gavin Newsom is going to bring you in.

I would imagine that's going to take, like, 50 years with their regulations.

Maybe. We'll see. We'll see. And do you think that hyperscaler demand—obviously, we were just on with Lisa Su talking about the demand for chips over the next couple of years—is correlated to the demand for data centers? Do you think that's actually going to play out the way all the public markets are projecting, or are we in 1999 peak, where everybody thinks fiber is going to be deployed all over the world? It turns out all those projections were totally off.

I think the important trend to watch is the capital investment that's happening and the term over which that's happening. So—

I thought Meta backed off on it a little bit. Didn't they, for a little bit, talk about deploying like crazy and then pull back? Although he's obviously spending a billion dollars on a chief AI scientist now.

Yeah. I think the investments they're making in people are actually rounding errors compared to the investments they're making in infrastructure. And I think that's something to appreciate in this moment in time. People are betting their entire balance sheets. These are the biggest and best balance sheets in the history of business, and they're betting their entire balance sheet on the future infrastructure that's going to power the modern economy.

And in the data centers in Texas, what's the limiting factor? Is it workforce to actually go build these things? Is it materials? Is it the cooling towers? Is it the chips? Is it the hyperscalers giving you the contracts? What's the limiting reagent?

Labor is definitely a major constraint. Like I said, we have about 4,000 people on-site every day. We're going to have multiple sites that are operating with thousands of folks basically building this infrastructure. So labor is definitely one of the big bottlenecks, and we think it's really important for America to make these massive investments in the workforce to really build the infrastructure of the future.

Do you think that requires some real reskilling, where people from oil and gas or construction have to go into totally net-new fields, or is it something where you guys are actually able to pull on preexisting talent pools pretty quickly?

Both. There's a lot of existing labor at that facility in Abilene, where we're actually pulling labor from all 50 states at this point, believe it or not. So—

Making it like a company town, importing people in.

Yeah. We have about 50% of the people from Texas, but we are importing a lot of labor to make the project happen.

Do you see the company starting to go more full-stack, beyond just the operations of the data centers? How do you think about it? You started off focused on energy arbitrage, now to data centers. Where do you see yourselves going over time?

Yeah. Crusoe is a vertically integrated AI infrastructure business. Data centers are a key component to that, and I think one of the most important pieces to build right now—and one of the hardest things to do at speed. But we also have this managed AI cloud services layer that enables innovators to build large-scale AI applications on the platform.

Makes sense. Well, yeah, Chase, thanks so much for joining us on stage, and appreciate the talk.

Yeah, thanks, D.

Okay, everybody, we have a real treat for you. Jensen Huang is here.

What's the story with the jacket? You have one of those. You have, like, 6.

I have something like 50 or 60 of them.

So you really?

Yeah.

Wow. What is that? Tom Ford?

I think so. This one is, I think so.

Yeah, that's nice. I like it. I tried that on. It was way too much money.

Well, you guys are all so fashionable.

Coming from you guys, it actually means something.

Yeah.

Oh, yeah. Oops. Look at you. Look at you.

Hey, we've been talking a lot about opportunity. You've talked—

So, mine is like a model.

He is. He is. Okay. Good idea. He's definitely in his head. He's like, “Is Tom Ford your favorite?” Who's your favorite, right?

My favorite is whatever my wife gets me.

Ah, she dresses you, right?

As soon as she gets it for me, it's my favorite.

Smart man. Nobody wears a suit better than Jacob. Good God.

Yeah, Jacob. He's a handsome man.

Just trying to keep up with you guys. I have 2 questions for you. Take them in whichever order you like.

We've been talking a lot about job displacement and opportunity, short-term and long-term. Obviously, you get to see everybody applying the technology because you've got the best product in town to build on. Therefore, everybody explains to you their hopes and their dreams. So, you have a unique way of looking at the playing field. You have complete information that we don't have. So, I want to know what you think. Don't worry, we'll fix it. What do you think about job creation, transfer, displacement, and so on? And then the second one, I've just always been curious.

You have all these important people knocking on your door. You have Zuck, you have Elon, you have Sam Altman. He seems like he's a little bit of a headache, I'll be honest.

He's great.

He's great. I'm joking. I'm joking. How do you allocate the H100s and whatever else you're selling them and still have them all like you? Because they must ask sometimes, “Hey, can I get extra? I'll pay you extra.” So, just the allocation of a finite amount of resources, and then jobs.

First of all, I wrote off $5 billion worth of Hoppers. If anybody would like to have some extras, you've got them. Just give me a call.

Jobs—we use AI across the whole company. Every single software engineer today uses AI. Not one left behind. 100% of our chip designers use AI. We are busier than ever. And the reason for that is because we have so many ideas that we want to go pursue. AI makes it possible for us to go pursue those ideas now that we're not doing the mundane stuff.

And so I think the first idea is, the more productive you are as a company, so long as you have more ideas, you can pursue those ideas. You'll go after those ideas.

And I think that AI, in my case, is creating jobs. It causes us to be able to create things that other people—customers—would like to buy. It drives more growth. It drives more jobs. All that goes together.

The other thing to remember is that AI is the greatest technology equalizer of all time.

Okay, explain.

Everybody's a programmer now.

Yes.

You used to have to know C, then C++ and Python, and in the future, everybody could program a computer, right? You just have to go up to the AI and say, “How do I program an AI?” The AI explains to you exactly how to program the AI.

Even when you're not sure exactly how to ask a question, you say, “What's the best way to ask the question?” It will actually write the question for you. It's incredible, and so it's a great equalizer.

Everybody is going to be augmented by AI. Everybody's an artist now. Everybody's an author now. Everybody's a programmer now. That is all true. We know that AI is a great equalizer.

We also know that although everybody's job will be different as a result of AI, some jobs will be obsolete, but many jobs will be created. The one thing that we know for certain is that if you're not using AI, you're going to lose your job to somebody who uses AI. That, I think, we know for certain.

There's not a software programmer in the future who's going to be able to hold their own typing by themselves.

Yeah. You can't raw-dog it. No.

Not anymore. No raw dog.

Not anymore. You can't raw-dog it. Tell us.

I'll be sure to go home and tell people.

Yeah. Exactly.

You're not going to raw-dog this.

Yeah. Get your Copilot on.

Now, what about the allocation?

Okay. So, the way we allocate is this.

The way we allocate is this: place a PO.

Okay, that's it. You go to the register, you pay, you order. First, in the old days with Hopper, it happened so fast it was impossible to keep up with the demand. But now we disclose our roadmap to all of our partners a year in advance. That gives everybody a chance to plan with us.

They decide how much power, how much data center space, and how much capex they want to allocate. We plan together. We work on transitions. It's really quite orderly these days.

What's the lifespan now? I was looking into how they're amortizing these units—4, 5 years. What happens to this massive buildout in years 6, 7, and 8? What will be the use of those computers if you keep building such great products that replace them at 2, 3, 4 times? What do we do with all that?

Two concepts are happening right now. The first thing is that every generation, we increase the performance by X factors.

If performance per dollar and performance per watt go up by X factors, whatever your data center power is, we just increase your revenues by X factors. Performance per watt is equal to revenues, and performance per dollar equals the cost. We increase your performance per dollar by X factors; we reduce your cost by X factors. Does that make sense? That's the first idea.

The reason why we're moving so fast is that we're trying to increase everybody's revenues and decrease everybody's costs, so that we have the benefit of driving AI costs down as far as possible, so that we can have thinking AI.

Right?

It's not that we're trying to make AI so that it generates 1,000 tokens and that's it. In the future, you're going to be generating millions of tokens, and it generates an answer as a result of that. You've got to think a long time, and so you've got to get that cost down.

The second idea is, if you look at the residual value of NVIDIA gear right now, Hopper, for example, one year later is probably about 75% to 80% of the original value. Then one year later, it's another 65%, and then one year later, it's about 50%.

The reason is that CUDA is so programmable, and we're constantly—the whole world, not just us, the whole world—is doing open-source development and improving its effectiveness. What's amazing is that the performance of Hopper increases over time because we're improving the software stack.

Hopper improved in performance by us and others by a factor of 4 since the time that we shipped it.

Right.

You can't get that out of a CPU, right?

Jensen, can you explain to us Elon's tweet and the impact on your industry? He said, “We're going to have 50 million H100 equivalents 5 years from now.” Everybody started to feverishly do the math, because if he has 50 million H100 equivalents, then OpenAI will have that much or more, Meta will have that much or more, Google, and so on.

Can you just explain to us, in layman's terms, what that means, what he just said, and how it impacts your business?

One of the biggest observations about AI is that there's an industry of applications that AI has created. It's a revolutionary technology; every industry will be revolutionized. New applications will be created, and so on and so forth.

All the things that we know—agentic AI, reasoning AI, robotics AI, and so on and so forth—we know all those things now. Every industry—health care, education, transportation, you name it, manufacturing—will be revolutionized.

The one part that we observed, and made a great contribution to, is that in order to sustain those applications, you need factories of AI. You have to produce AI. Unlike software, you write the software and that's it. In the case of AI, you have to continuously produce it—generate the tokens.

Right.

In a lot of the same ways that energy production was a large part of the economy 200 or 300 years ago, I think it actually peaked at 30%.

Yep.

There's going to be a whole industry just producing tokens, and this is going to be the new infrastructure. Just as we have the energy production infrastructure, we have the internet infrastructure, and we've got to build out that plumbing. Now we have to build out the AI infrastructure.

My sense is that we're probably a couple of hundred billion dollars, maybe a few hundred billion dollars, into a multitrillion-dollar infrastructure buildout per year.

Yeah. What about manufacturing?

The reason for that is because you want the new infrastructure, which increases revenue, driving your costs down.

Right.

That's right.

What about manufacturing in the U.S.? Where are we? We've seen stories of TSMC in Arizona. We asked this question earlier about how it's going. Is the U.S. equipped? What is it going to take for us to get there and have onshore fabs?

First of all, you guys know you're talking about the United States. I know that there are lots of concerns, and everybody's worried about competition and things like that, but we are talking about America here. This is unquestionably the most technology-rich country in the world, and this is the most innovative country in the world.

The computer industry I have the honor to serve is the single greatest industry our country has ever produced. I think we could acknowledge that.

Yep.

The level of leadership of the computer industry, the technology industry, is just unimaginable worldwide. This is our national treasure. This is one of our country's assets. We have to make sure that we continue to advance it.

Onshoring next-generation manufacturing is going to be insanely technology-driven: robotics technology, AI technology. You're going to have factories orchestrated by AI, orchestrating a whole bunch of robots that are AI, building products that are effectively AIs, right? You're going to have these layers of inception. The amount of technology necessary to create that is really insane.

I love President Trump's bold vision of reindustrializing the United States. That entire band of industry that's missing—we outsourced too much of it. Frankly, we don't need to insource all of it, but we ought to bring onshore the most advanced, the most economy-sustaining and driving, national-security-enhancing parts of the industry.

People always degrade it down to tennis shoes. We don't have to go there. We just manufacture chips and AI supercomputers. In Arizona and Texas, we will, in the next 4 years, probably produce about half a trillion dollars' worth of AI supercomputers. That half a trillion dollars' worth of AI supercomputers will probably drive a few trillion dollars' worth of AI industry, right?

And so that's only in the next several years, and they're doing great. Arizona is doing great.

And so there's a lot of talk about American competitiveness today. The White House rolled out its AI Action Plan, and NVIDIA is making very big bets on the United States. As the CEO of a global company, what do you see as America's unique advantages that other countries don't have?

America's unique advantage that no country possibly has is President Trump. Let me explain why.

One, on the first day of his administration, he realized the importance of AI, and he realized the importance of energy. For the last—I don't know how many years—energy production was vilified, if you guys remember.

Yeah.

You can't sustain a brand-new industry like artificial intelligence without energy. If we decide as a country that the only thing we want is IP, to be an IP-only, services-only country, then we don't need much energy. But if we want to produce things—something as vital as artificial intelligence—we need energy. And so I'm just delighted to see President Trump accelerate AI innovation and accelerate the growth of energy so that we can sustain this new industry and go after the new industrial revolution. Big, big deal.

Can you talk about physical AI versus data-center AI? We talked a little bit about this today. Is there a threshold where you see physical AI accelerating and ultimately the deployment of chips outpaces the deployment of chips in data centers? Is that where the world evolves to, or what do you think the world looks like?

Excellent. Everything in the world that moves will be autonomous someday, and that someday is probably around the corner. Everything that moves—we already know that your lawn mower is going to be autonomous. Who's going to be pushing a lawn mower around? That's craziness, unless you want to.

I think everything that moves will be autonomous, and every machine—every company that builds machines—will have 2 factories. There's the machine factory, for example, for cars, and then there's the AI factory to create the AI for the cars.

Maybe you're a machine factory building humanoid robots. You need an AI factory to build a brain for the humanoid robot, right? And so every company in the future—in fact, the future of industry—is really 2 factories.

Tesla already has 2 factories, right? Elon has a giant AI factory. He was very early in recognizing that he needs to have an AI factory to sustain the cars that he has now. Now he's got AI in the car, but in the future, instead of a whole lot of people remotely monitoring and controlling it, it'll be a giant AI that's doing the remote control. Then, only in the case that the giant AI can't handle it, will a person come in to intercept.

I think you see that these industries in the future—every industrial company will be an AI company, or you're not going to be an industrial company.

There were a couple of moments throughout the course of this year where people almost threw in the towel and said, “Oh, we lost to China.” There was the DeepSeek moment, and then maybe this week or last week, there was this Kimi model moment, but then it kind of fizzled out. Can you explain to us how big of a threat they really are in terms of getting to supremacy, getting there first, whether it's AGI or superintelligence?

Excellent question. The Chinese AI labs are the world's leading open-model companies. They offer the most advanced open models. Open source is fantastic. If not for open source, we know startups won't exist. To the extent that we believe the future industry is going to be today's startups, they're going to need open-source models.

DeepSeek, when it came out, was a great win for the United States. It was an incredible win. What people didn't realize—there are 2 reasons. First, imagine if DeepSeek came out and only ran on Huawei. I just want us to pretend. Use that thought experiment.

Totally. You've got 2 parallel universes.

Exactly. Could you imagine if Qwen came out and only worked on a non-American tech stack? Could you imagine if Kimi came out and only worked on non-American technology? These are the top 3 open models in the world today. They are downloaded hundreds of millions of times.

The fact of the matter is that the American tech stack all over the world, being the world's standard, is vital to the future of winning the AI race. You can't do it any other way. We've got to be—as you know, any computing platform wins because of developers.

Yeah. And half of the world's developers are in China.

The second thing—and it's really a big deal—is that when DeepSeek came out, we were thrilled for the second reason, which is that we now have a super-efficient reasoning model. The reason for that is because the old models are one-shot: Give it a question, and everything was memorized. Pretraining is basically memorization and generalization—2 concepts. Post-training is teaching you how to think.

And so now, with DeepSeek R1, Kimi K2, and Q13, you have reasoning models that can help you think. The reason why I was so excited is, if each pass of a thought is energy-efficient, then you can think for a long time, right?

Yeah. The last question from me is that we see this capital being applied to human capital in a way that we never thought was possible. It used to be NBA players signing $300 million contracts; now it's model researchers. There was a post this weekend that said there was a person who was offered $1 billion over 4 years by Meta.

If that's happening at this layer, why hasn't it happened at your layer? Because you are the enabler of all of that. How do you think all of this human capital is actually going to play out?

First of all, I've created more billionaires on my management team than any CEO in the world. They're doing just fine, so don't feel sad for anybody at my layer.

Yeah, everybody's doing okay.

Yeah, my layer is doing just fine. I tell you, the important—the big idea, though, is what you're highlighting: The impact of 150 or so AI researchers can, with enough funding behind them, create an OpenAI.

150 people?

Yeah. Well, DeepSeek has 150 people. Moonshot has 150 people, right? I mean, look at the original OpenAI: It was about 150 people. DeepMind, you know, they're all about that size.

I think there's something about the elegance of small teams, and that's not a small team. That's a good-sized team with the right infrastructure. And so that kind of tells you something: If you're willing to pay, say, $20 billion or $30 billion to buy a startup with 150 AI researchers, why wouldn't you pay $1 billion?

Right. By the way, we need to wrap because I'm going to do this one question. Somebody who was inside your organization told me that, with the options, you have a secret pool of options and that, if somebody does a great job, you will randomly drop a bunch of RSUs on top of them. You have this little bag of options you carry around and hand them out. Is that true?

That's nuts.

Is that true?

Yeah, I'm carrying them in my pocket right now. So listen, this is what happens: I review everybody's compensation. At the end of every cycle, when they present it, they send me everybody's recommended comp.

I go through the whole company. I've got my methods of doing that, and I use machine learning. I use all kinds of technology, and I sort through all 42,000 employees. 100% of the time, I increase the company's spend on OpEx, and the reason for that is because you take care of people. Everything else takes care of itself.

All right. Well done. Thank you.

Thank you, James. Great to see you.

Great to see you. We have an event in LA. We'd love to continue the conversation, so we'll send you a note.

The world's number 1 podcast.

There you go. Thank you.