NVIDIA：OpenAI、算力的未来与美国梦｜BG2 与 Bill Gurley、Brad Gerstner 对谈

I think that OpenAI is likely to be the next multi-trillion-dollar hyperscale company.

Jensen, great to be back, of course, with my partner, Bill Gurley.

Welcome to NVIDIA. Oh, and nice glasses.

Those actually look really good on you. The problem is now everybody’s going to want you to wear them all the time. They’re going to say, “Where are the red glasses?”

Bill Gurley

I can vouch for that.

1. The Year in AI Recap

It’s been over a year since we did the last pod.

Yeah.

Over 40% of your revenue today is inference, but inference is about to go up because of chain-of-reasoning.

Right.

It’s about to go up by a billion times, right? By a million X, by a billion.

That’s right. That’s the part that most people haven’t completely internalized. This is that industry we were talking about. This is the Industrial Revolution.

Honestly, it’s felt like you and I have had a continuation of the pod every day since then. In AI time, it’s been about 100 years. I was rewatching the pod recently, and the many things that we talked about stood out. The one that was probably most profound for me was you pounding the table that, remember, at the time there was kind of a slump in terms of pretraining, and people were like, “Oh, my God, the end of pretraining. The end of pretraining. We’re not going. We’re overbuilding.” This was about a year and a half ago.

You said inference isn’t going to go 100X or 1,000X; it’s going to go 1 billion X.

Mm-hmm.

I underestimated. Let me just go on record. I estimated we now have three scaling laws, right? We have a pre-training scaling law. We have a post-training scaling law. Post-training is basically like AI practicing—practicing a skill until it gets it right. It tries a whole bunch of different ways, and in order to do that, you've got to do inference. So now training and inference are integrated in reinforcement learning. That's called post-training. And then the third is inference. The old way of doing inference was one-shot, right? But the new way of doing inference, which we appreciate, is thinking. So think before you answer. The longer you think, the better the quality answer you get. While you're thinking, you do research, you go check on some ground truth. You learn some things, you think some more, you go learn some more, and then you generate an answer. Don't just generate right off the bat.

Which brings us to where we are today. You knew that last year, but is your level of confidence this year that inference is going to go 1 billion X—and where that will take the levels of intelligence—is it higher? Are you more confident this year than you were a year ago?

I’m more confident this year, and the reason for that is, look at the agent systems now. AI is no longer a language model. AI is a system of language models, and they’re all running concurrently, maybe using tools. Some of them are using tools, some of them are doing research, and there’s a whole bunch of stuff. It’s all multimodality, and look at all the video that’s being generated. It’s just crazy stuff.

2. OpenAI Stargate & Nvidia Investment

It really brings us to the seminal moment this week that everybody’s talking about: the massive deal you announced a couple of days ago with OpenAI and Stargate, where you’re going to be a preferred partner and invest $100 billion in the company over a period of time. They’re going to build 10 gigawatts, and if they use NVIDIA for those 10 gigawatts, that could be upwards of $400 billion in revenue to NVIDIA. Help us understand—tell us a little bit about that partnership, what it means to you, and why that investment makes so much sense for NVIDIA.

I’ll answer that last question first, and then I’ll come back and explain my way through it. I think that OpenAI is likely going to be the next multi-trillion-dollar hyperscale company.

Okay. Why do you call it a hyperscale company?

Hyperscale—like Meta is hyperscale. Google is hyperscale. They’re going to have consumer and enterprise services, and they are very likely going to be the world’s next multi-trillion-dollar hyperscale company. I think you would agree with that.

I agree.

If that’s the case, the opportunity to invest before they get there is one of the smartest investments we can possibly imagine. You’ve got to invest in things you know, right? It turns out we happen to know this space, and so the opportunity to invest in that—the return on that money is going to be fantastic. We love the opportunity to invest. We don’t have to invest, right? It’s not required for us to invest, but they’re giving us the opportunity to invest. It’s a fantastic thing.

Now, let me start from the beginning. We’re partnering with OpenAI in several projects. The first project is the build-out of Microsoft Azure. We’re going to continue to do that, and that partnership is going fantastically. We have several years of build-out to do—hundreds of billions of dollars of work just there.

Right.

The second is the OCI build-out. I think there are 5, 6, 7 gigawatts that are about to be built out, and so we’re working with OCI, OpenAI, and SoftBank to build that out.

Right.

Those projects are contracted. We’re working on them. There’s lots of work to do. Then the third is CoreWeave, right?

I’m talking about OpenAI still.

Yes.

Okay. Everything in the context of OpenAI.

The question is, what is this new partnership? This new partnership is about helping OpenAI—partnering with OpenAI—to build its own self-built AI infrastructure for the first time.

Right.

This is us working directly with OpenAI at the chip level, at the software level, at the systems level, and at the AI factory level to help them become a fully operational hyperscale company. This is going to go on for some time, and it’s going to supplement the amount of compute they’re going through. They’re going through 2 exponentials, as you know.

Right.

The first exponential is the number of customers, which is growing exponentially. The reason for that is the AI is getting better, the use cases are getting better, and just about every application is connected to OpenAI now. They’re going through the usage exponential.

The second exponential is the computational exponential of every use.

Yes.

3. Future of ASICs & Economics

Instead of just a one-shot inference, it now thinks before it answers. These 2 exponentials are compounding their compute requirements, and so we’ve got to build out all these different projects. This last one is additive on top of everything that they’ve already announced and all the things that we’re already working on with them. It’s additive on top of that, and it’s going to support this incredible exponential growth.

One of the things you said there that’s really interesting to me is that they’re going to be a high-probability, multi-trillion-dollar company in your mind. You think it’s a great investment. At the same time, they’re self-building. You’re helping them self-build their data centers. Heretofore, they’ve been outsourcing to Microsoft to build the data centers. Now they want to build full-stack factories themselves.

They want to basically have a relationship with us the way that Elon and xAI have a relationship with us.

Correct.

I mean, Elon and xAI built—

Exactly. But I think this is a very big deal when you think about the advantage that Colossus had. They’re building full-stack. That is a hyperscaler, because if they don’t use the capacity, they could sell it to somebody else. In the same way, Stargate is building monstrous capacity. They think they’ll need to use most of it, but it puts them in a position to sell it to somebody else as well. It sounds very much like AWS, GCP, or Azure. That’s what you’re saying.

I think they’ll likely use it themselves. Just like in the case of X, they’ll likely use it themselves. But they would like to have the same direct relationship with us—a direct working relationship and a direct purchasing relationship.

Meta, just as Zuck and Meta have with us, has exactly a direct relationship with us. Our relationship between us and Sundar and Google is direct. Our partnership with Satya and Azure is direct. Isn’t that right?

They’ve gotten to a large enough scale that they believe it’s time for them to start building these direct relationships. I’m delighted to support that. Satya knows it, Larry knows it, and everybody is aware of what’s going on. Everybody is very supportive of it.

4. Nvidia Accelerated Compute TAM

One of the things I find mysterious: you just mentioned Oracle—$300 billion—Colossus and what they’re building. We know what the sovereigns are building. We know what the hyperscalers are building. Sam is talking in terms of trillions. But of the 25 sell-side analysts on Wall Street who cover your stock, if I look at the consensus estimate, it basically has your growth flatlining starting in 2027.

8% growth from 2027 through 2030. Okay, that is the 25 people and their only job. They get paid to forecast the growth rate for NVIDIA.

We're comfortable with that, by the way.

Right.

Look, we're comfortable with that. We have no trouble beating the numbers on a regular basis, right?

No, I understand that. But there is this interesting disconnect. I hear it every day on CNBC and Bloomberg, and I think it goes to some of these questions around shortages leading to a glut that they don't believe. They say, “Okay, we'll give you credit for 2026, but 2027, maybe we'll have too much and you're not going to need that.”

It is interesting to me, and I think it's important to point out, that your consensus forecast is that this won't happen, right? We also put together forecasts for the company, taking into account all of these numbers, and what it shows me is that, even though we're 2½ years into the age of AI, there's still a massive divergence of belief between what we hear Sam Altman saying, you saying, Sundar Pichai saying, and Satya Nadella saying, and what Wall Street still believes. Again, you're comfortable with that.

I also don't think it's inconsistent.

Okay, so explain that a little bit.

First of all, for the builders, we're supposed to be building for opportunity, right? We're builders. Let me give you 3 points to think through, and these 3 points will hopefully help you be more comfortable with NVIDIA in this future. The first point—and this is the laws-of-physics point—is the most important point: general-purpose computing is over, and the future is accelerated computing and AI computing.

Right.

The way to think about that is: how many trillions of dollars of computing infrastructure in the world has to be refreshed?

Right. Right.

When it gets refreshed, it's going to be accelerated computing.

That's right.

The first thing you have to realize is that general-purpose computing is going to go to accelerated computing, and nobody disputes that. Everybody says, “Yeah, we completely agree with that. General-purpose computing is over. Moore's law is dead.” People say these things.

Our partnership with Intel recognizes that general-purpose computing needs to be fused with accelerated computing to create opportunities for them.

Is that right?

And so, 1, general-purpose computing is shifting to accelerated computing and AI. 2, the first use case of AI is actually already everywhere, right? It's in search, recommender engines, and shopping.

Basic hyperscale computing infrastructure used to be CPUs doing recommenders, right? It's now going to be GPUs doing AI, right? You just take classical computing, and it's going to accelerated computing and AI. You take hyperscale computing from CPUs to accelerated computing and AI.

That means feeding Meta, Google, ByteDance, and Amazon, and taking their classical, traditional way of doing hyperscale computing and moving it into AI. That's hundreds of billions of dollars. There may be 4 billion people on the planet today—if you take TikTok, Meta, Google, and Amazon into account—who are already demanding workloads driven by accelerated computing.

That's exactly right.

And so there are simply new opportunities without even thinking about AI creating new opportunities. It's about AI shifting how you used to do something to the new way of doing something.

Okay, and then now let's talk about the future. So far, I've only spoken largely about just mundane stuff. The old way is now wrong. You're no longer going to use fuel-lit lanterns; you're going to go to electricity. That's all. You no longer use prop planes; you're going to go to jets. That's all.

Right.

So far, that's all I've talked about. Now, the incredible thing is, when you go to AI, when you go to accelerated computing, what happens? What are the new applications that emerge as a result? That's all the AI stuff that we're talking about, and that's the opportunity.

What is it? How does that look? The simple way of thinking about that is that where motors replaced labor and physical activity, we now have AI. These AI supercomputers, these AI factories that I talk about, are going to generate tokens to augment human intelligence, right?

Human intelligence represents, what, 55% to 65% of the world's GDP. Let's call it $50 trillion. That $50 trillion is going to get augmented by something.

Yeah.

Let's come back to a single person. Suppose I were to hire a $100,000 employee and augment that $100,000 employee with a $10,000 AI.

Yes.

If that $10,000 AI, as a result, made that $100,000 employee twice as productive or 3 times more productive, would I do it? In a heartbeat. I'm doing it across every single person in our company right now.

Every single co-agent?

That's right. Every single software engineer, every single chip designer in our company already has AIs working with them.

100% coverage.

As a result, the number of chips we're building is bigger. The number is growing, the pace at which we're doing it is faster, and so we're growing faster as a company. As a result, we're hiring more people. Our productivity is greater, our top line is greater, and our profitability is greater. What's not to love about that?

Now apply the NVIDIA story to the world's GDP.

Yeah.

What's likely to happen is that that $50 trillion is augmented by, let's pick a number, $10 trillion. That $10 trillion needs to run on a machine.

The reason that AI is different from the past is that, in a way, software was written a priori, and then it ran on a CPU. It ran when a person would operate it. In the future, of course, AI is generating tokens, but a machine has to generate the tokens, and it's thinking.

That software is running all the time, whereas in the past the software was written once. Now the software is, in fact, writing all the time. It's thinking. In order for the AI to think, it needs a factory.

Let's say that $10 trillion of token generation has 50% gross margins, and $5 trillion of it needs a factory—it needs AI infrastructure. If you told me that, on an annual basis, the capex of the world was about $5 trillion, I would say the math seems to make sense.

Yeah.

That's kind of the future, right? Going from x86 general-purpose computing to accelerated computing, replacing all the hyperscale computing with AI, and then augmenting human intelligence for the world's GDP. Today, that market—our estimate—is about $400 billion annually.

Yeah.

The TAM is a 4–5× increase over where it is today.

Bill Gurley

Yeah. Eddie Wu at Alibaba said last night that between now and the end of the decade, they're going to increase their data-center power by 10×.

Right. Right. You just said how much?

Bill Gurley

4×.

There you go.

Bill Gurley

Yeah.

They're going to increase power by 10×, and we correlate to power. NVIDIA's revenue is almost correlated to power, isn't that right?

Bill Gurley

Yeah.

That's right. Yeah, because—

Bill Gurley

One other thing—what else did he say?

Yeah. He said token generation is doubling every few months.

Bill Gurley

Yeah.

What's that saying? Performance per watt has to keep going exponentially. That's why NVIDIA is cranking it out with performance per watt. Revenue per watt is basically revenue in this future.

Bill Gurley

Embedded in this assumption, I find it very fascinating from a historical context. For 2,000 years, basically, GDP did not grow. Then we get the Industrial Revolution, and GDP accelerates. We get the digital revolution, and GDP accelerates.

Basically, what you're saying—and Scott Bessent has said it; he said, “I think we're going to have 4% GDP growth next year”—is that the world's GDP growth is going to accelerate because now we're giving the world billions of co-workers that will do work for us. If GDP is an amount of output for a fixed amount of labor and capital, it has to accelerate.

It has to, right? Look at what's going on with AI as a result of the technology of AI. That technology—let's just call it the large language models and all the AI agents—is now creating a new industry of AI agents. There's no question about that.

OpenAI is the fastest-growing revenue company in history, right? They're growing exponentially. AI itself is a fast-growing industry because AI needs a factory behind it, right?

An infrastructure behind it.

This industry is growing. My industry is growing, and because my industry is growing, the industry underneath it is growing. Energy is growing, and power generation is growing. This is like a renaissance for the energy industry, isn't that right? Nuclear energy, gas turbines—look at all of those companies in the infrastructure ecosystem underneath us. They're doing incredibly well.

5. NVDA ROI – Glut or Bubble?

Everybody's growing. These numbers have everybody talking about a glut bubble, right? Zuckerberg said last week on a podcast, “Listen, I think it's quite possible at some point that we will have an air pocket, and Meta may, in fact, overspend by $10 billion or whatever.” But he said it doesn't matter. It's so existential to the future of his business that it's a risk they have to take. When you think about that, it sounds a little bit like the prisoner's dilemma, right?

These are very happy prisoners.

Walk us again through that.

Today, our estimate is that we're going to have $100 billion of AI revenue in 2026, excluding Meta and excluding the GPUs running recommender engines.

Or search?

Correct. So there's other stuff, but let's call it $100 billion.

What is that industry, anyway?

The industry is already in hyperscale. What is hyperscale? You know, it's in the trillions. By the way, that industry is going to AI. Before anybody starts at zero, you've got to start there.

But I think the skeptics would say we need to go from $100 billion of AI revenue in 2026 to at least $1 trillion of AI revenue in 2030. You were just talking a minute ago about $5 trillion when you look at global GDP. If you did a bottoms-up analysis, can you see your way to $1 trillion of AI-driven revenues from $100 billion over the course of the next 5 years? Are we growing that fast?

Yes. I would also say we're already there.

Okay, so explain that.

The hyperscalers went from CPUs to AI. Their entire revenue base is now AI-driven. You can't do TikTok without AI.

Correct.

You can't do YouTube Shorts without AI. You can't do any of this stuff without AI. Look at the amazing things Meta is doing with customized and personalized content. You can't do that without AI. All of that used to be humans creating content a priori, creating 4 choices that were then selected by a recommender engine. Now it's an infinite number of choices generated by an AI, right?

But those things are already—like, we had the transition from CPUs to GPUs largely for those recommender engines.

And that's fairly new, I would say—within the last 3 or 4 years. Zuck would tell you—I was at SIGGRAPH, and Zuck would tell you they were late getting to GPUs, for sure.

GPUs for Meta are what, 2 and a half years?

It's pretty new. Search with GPUs—

For sure.

Brand-spanking new.

For sure, for sure.

Brand-spanking new. Search on GPUs.

So your argument would be that the probability we're going to have $1 trillion of AI revenues by 2030 is near certain because we're almost already there.

Let's just talk about the incremental from where we are.

Now we can talk about the incremental from where we are today, right? As you do your bottoms-up or your tops-down, I just heard your top-down about the percentage of global GDP.

Yeah.

What is the percentage probability that you think we'll run into a glut in the next 3, 4, or 5 years?

Right. It's a distribution. We don't know the future. It's a distribution of power. Until we fully convert all general-purpose computing to accelerated computing and AI, until we do that—

Yes.

—I think the chances are extremely low.

Okay, okay. And that will take a few years.

That'll take a few years.

Yeah. Let me ask one more, and then—

Until all recommender engines are AI-based, until all content generation is AI-based—because consumer-oriented content generation is very largely recommender systems and so on and so forth—and all of that's going to be AI-generated. Until all of the stuff that classically was hyperscale now transitions to AI, everything from shopping to e-commerce—all that stuff—until everything goes over.

But all this new build, right? When we're talking about trillions, we're investing ahead of where we are. Is that at will? Are you obliged to invest the money even if you see a slowdown or a glut coming? Or is this one of those things where you're waving the flag to the ecosystem to say, “Get out and build,” and at some point, if we see some of this slow down, we can always pull back on the level of investment?

Actually, it's the other way, because we're at the end of the supply chain, right? We respond to demand. Right now, all the VCs will tell you—and you guys know—the demand is strong. There's a shortage of compute in the world, not because there's a shortage of GPUs in the world.

Okay.

If they give me an order, I'll build it.

Mhm.

Over the last couple of years, we've really plumbed the supply chain. All of the supply chain behind me—from wafer starts to CoWoS and HBM memory, all of that technology—we've really geared up.

Yeah.

If we need to double, we'll double.

Yes.

The supply chain is ready. Now we're just waiting for demand signals. When the CSPs and the hyperscalers and our customers do their annual plan and give us their forecast, we respond to that and build to it.

Now, what's going on, of course, is that every one of their forecasts that they provide us turns out to have been wrong—

Right.

—because they underforecasted. Now we're always in scramble mode.

Mhm.

We've been in scramble mode now for a couple of years, and whatever forecast we've been given has always been a significant increase from last year, but not enough.

Satya last year seemed to be pulling back a little bit. Some people called him the adult in the room, tamping down some of these expectations. A few weeks ago, he said, “Hey, I've also built 2 gigawatts this year, and we're going to accelerate in the future.” Do you see some of the traditional hyperscalers that may have been moving a little slower than, let's call it, CoreWeave or Elon xAI, or maybe a little slower than Stargate? Do you see them all leaning in more now? It sounds to me like they're all leaning in more now, and they're all also—

Because of the second exponential.

Okay.

We've already had 1 exponential that we were experiencing, which was the adoption rate of AI. The engagement with AI was growing exponentially.

Yes.

The second exponential that kicked in was reasoning.

Yeah. That was the conversation we had 1 year ago.

1 year ago.

Yeah. We said, “Hey, listen. The moment you take AI from 1-shot—memorizing an answer—”

Right. Memorizing and generalizing, that's basically pretraining.

Yeah.

Memorizing an answer—what's 8 × 8? Just memorize it. So memorizing an answer and generalizing, that was 1-shot AI. Now, 1 year ago, reasoning came about—

For sure.

Research came about, tool use came about, and now you're a thinking AI.

1 billion X.

It's going to use a lot more compute. Certain hyperscaler customers, to your point, had internal workloads that they had to migrate anyway from general-purpose computing to accelerated computing. So they built through the cycle. I think maybe some hyperscalers had different workloads, so they weren't quite sure how quickly they could digest it, but everyone has now concluded that they dramatically underbuilt.

One of the applications that I favor is just good old-fashioned data processing: structured data and unstructured data. Just good old-fashioned data processing. Very soon, we're going to announce a very big initiative in accelerated data processing.

Data processing represents the vast majority of the world's CPUs today. It still completely runs on CPUs. If you go to Databricks, it's mostly CPUs. You go to Snowflake, it's mostly CPUs. SQL processing at Oracle is mostly CPUs. Everybody's using CPUs to do SQL and structured data.

In the future, that's all going to move to AI data processing. That is one gigantic, massive market that we're going to move to. But everything NVIDIA does requires acceleration layers and domain-specific data-processing recipes. We've got to go build that, but that's coming.

6. Roundtripping Claims

Bill Gurley

One of the pushbacks—I turned on CNBC yesterday, and they were talking about a glut bubble. When I turned on Bloomberg, it was about round-tripping and circular revenues. For the benefit of people at home, these arrangements are when companies enter into a misleading transaction that artificially inflates revenue without any underlying economic substance. In other words, growth propped up by financial engineering, not by customer demand. The canonical case everybody's referencing, of course, is Cisco and Nortel from the last bubble 25 years ago.

When you guys, Microsoft, or Amazon are investing in companies that are also your big customers—in this case, you guys investing in OpenAI while OpenAI is buying tens of billions of dollars of chips—remind us, and remind everybody else, what are the analysts on Bloomberg and elsewhere getting wrong when they're hyperventilating about circular revenues or round-tripping?

10 gigawatts is like $400 billion, right? Something like that.

And that $400 billion will have to be largely funded by their offtake, right? Their revenue is growing exponentially. It has to be funded by their capital—the money they've raised through equity and whatever debt they can raise. Those are the 3 vehicles.

The equity that they could raise and the debt that they could raise have something to do with the confidence in the revenues that they could sustain, for sure. Smart investors and smart lenders will consider all of these factors. Fundamentally, that's what they're going to do. That's their company. It's not my business.

Of course, we have to stay very close to them to make sure that we build in support of their continued growth.

Okay. So there's the revenue side of it, and it has nothing to do with the investment side of it. The investment side of it is not tied to anything. It's an opportunity to invest in them. As we were mentioning earlier, this is likely going to be the next multitrillion-dollar hyperscale company. Who doesn't want to be an investor in that?

My only regret is that they invited us to invest early on. I remember those conversations, and we were so poor—we didn't invest enough. I should have given them all my money.

The reality is, if you guys don't do your jobs and keep up—if Vera Rubin doesn't turn into a good chip—they can go get other chips and put them in these data centers, right? There's no obligation that they have to use your chips. Like you said, you're looking at this as an opportunistic equity investment.

The other thing I would say—and we've made some great investments. I've got to put it out there—we invested in xAI, and we invested in CoreWeave.

Incredible. Yeah.

Yeah. How smart was that?

Yeah.

As I go back to this, the other fundamental thing it seems to me is that you're putting it out there. You're saying, “This is what we're doing.” The underlying economic substance here is not that you're somehow just sending revenues back and forth between the 2 companies. We've got people sending money every month for ChatGPT, with 1.5 billion monthly users using the product.

You just said every enterprise in the world is either going to do this or they will die. Every sovereign views this as existential to their national security and economic security, as nuclear power. What person, company, or nation says intelligence is basically optional for us? I mean, it's fundamental to them.

7. Annual Release Cadence & Extreme Co-design

Well, the automation of intelligence—

I beat the demand question to death. So let's jump in a little bit to system design. I'm going to turn to Bill here in a second on that.

But in 2024, you switched to your annual release cycle, right, with Hopper. You then had a massive upgrade, which required significant data-center overhaul, with Grace Blackwell. In 2025, and in the back half of 2026, we're going to get Vera Rubin. In 2027, we'll get Ultra, and in 2028, Feynman.

How is the annual release cycle going? What were the main goals of going to an annual release cycle? And did AI inside NVIDIA allow you to execute the annual release cycle?

Yeah, the answer is yes. On the last question, without it, NVIDIA's velocity—our pace, our scale—would be limited. Without AI these days, it's simply not possible to build what we've built.

Why do we do it? Remember, Eddie Wu said it at his earnings call or his conference. Satya has said it, and Sam has said it. The token-generation rate is going up exponentially.

The customer use is going up exponentially. I think they're at 800 million weekly active users or something like that. Yes, I mean, that's less than 2 years from ChatGPT, right? Each of those users is generating massively more tokens because they're using inference-time reasoning.

That's right. Exactly.

The first thing is, because the token-generation rate is going up so incredibly—2 exponentials on top of each other—we have to increase performance at incredible rates. Otherwise, the cost of token generation will keep growing because Moore's law is dead, right? Transistors basically cost the same every single year now, and power is largely the same.

Between those 2 fundamental laws, unless we come up with new technologies to drive the cost down, even if there's a slight difference in growth and you give somebody a discount of a few percent, how is that going to make up for 2 exponentials? We have to increase our performance annually at a pace that keeps up with that exponential.

In the case of going from Kepler all the way to Hopper, it was probably 100,000x. That was the beginning of the AI journey for NVIDIA—100,000x in 10 years. Between Hopper and Blackwell, because of NVL72, we increased 30x in 1 year.

And then we'll get another X factor again with Rubin, and then we'll get another X factor with Feynman.

The way we do that is because the transistors aren't really helping us very much, right? Moore's law is largely about density growing, but performance is not. If that's the case, one of the challenges that we have is that we have to break the entire problem down at the system level and change every chip at the same time, along with the entire software stack and all the systems, all at the same time.

It's the ultimate extreme co-design. Nobody's ever co-designed at this level before, right? We change the CPU, revolutionize the CPU and the GPU, the networking chip, NVLink scale-up, and Spectrum-X scale-out.

Somebody said, “Oh, yeah, it's just Ethernet.” Yeah, right. Okay. Spectrum-X Ethernet is not just Ethernet. People are starting to discover, “Oh my God, the X factors are pretty incredible,” right?

NVIDIA's Ethernet business—the just-Ethernet business—is the fastest-growing Ethernet business in the world. Scale out, and of course now we have to build even larger systems. We scale across multiple AI factories connected together.

We do this at an annual pace. We now have an exponential of exponentials going on from our technology, and that allows our customers to drive the cost of tokens down, keep making those tokens smarter and smarter with pre-training and post-training and thinking.

As a result, when the AI gets smarter, it gets more use. When it gets more use, it's going to grow exponentially.

Bill Gurley

For people who may not be as familiar, what is extreme co-design?

Extreme co-design means that you have to optimize the model, algorithm, system, and chip at the same time. You have to innovate outside the box, right?

Moore's law said you just have to keep making the CPU faster and faster. Everything got faster. You were innovating within a box—just make that chip faster. Well, if that chip doesn't go any faster, what are you going to do? Innovate outside the box.

NVIDIA really changed things because we did 2 things: We invented CUDA, invented GPUs, and we invented the idea of co-design at a very large scale. That's why there are all these industries we're in. We're creating all these libraries and doing co-design.

Full-stack extreme is even beyond software and GPUs. It's now at the data-center level: switches and networking, all of that software in the switches and the networking and the NICs, the scale-up, the scale-out—optimizing across all of that.

As a result, Hopper to Blackwell is 30x. No Moore's law could possibly achieve that, right? That's extreme, and that comes from extreme co-design.

That's why NVIDIA got into networking and switching, scale-up and scale-out and scale-across, and building CPUs, GPUs, and NICs. That's the reason why NVIDIA is so rich in software and people. We check in more open-source software than just about anybody else in the world, except 1 other company. I think it's AI2 or something like that.

We have such enormous richness of software, and that's just in AI. Don't forget computer graphics, digital biology, autonomous vehicles, and all of that. The amount of software we produce as a company is incredible. That allows us to do deep and extreme co-designs.

Bill Gurley

I heard from one of your competitors, “Yes, he's doing this because it helps drive down the cost of token generation.” But at the same time, your annual release cycle makes it almost impossible for your competitors to keep up. The supply chain gets locked up more because you're giving 3-year visibility to your supply chain, so now the supply chain has confidence as to what they can build to.

Do you think about this?

Wait, wait, wait. Before you ask the question, think about this. In order for us to do several hundred billion dollars a year of AI infrastructure buildout—

Bill Gurley

Yes.

Think about how much capacity we had to go start a year ago.

Bill Gurley

Yes. We're talking about building hundreds of billions of dollars of wafer starts and DRAM buys.

Yeah. This is now at a scale that hardly any company can keep up with.

Bill Gurley

So would you say your competitive moat is greater today than it was 3 years ago?

Yeah.

First of all, there’s just more competition than ever before, but it’s harder than ever before. The reason why I say that is because wafer costs are getting higher, which means that unless you do co-design at an extreme scale, you’re just not going to be able to deliver the X-factor growth. Number one, unless you’re working on 6, 7, 8 chips a year, right? That’s the amazing thing: it’s not about building an ASIC; it’s about building an AI factory system.

And this system has a lot of chips in it, and they’re all co-designed. Together, they deliver that 10x factor that we get almost regularly. Okay, so number one, the co-design is extreme. The second thing is that the scale is extreme.

When your customers deploy a gigawatt, that’s 400,000, 500,000 GPUs, right? Getting 500,000 GPUs to work together is a miracle. It’s just a miracle. Your customers are taking enormous risk on you to go buy all of this. You have to ask yourself: What customer would place a $50 billion PO on an architecture, right? On an unproven architecture, a new one, right?

You just put out a whole new chip. You’re as excited as you are about it, and everybody’s excited for you, and you just show the first silicon, right? Who’s going to give you a $50 billion PO, right? And why would you start $50 billion worth of wafers for a chip that just taped out?

For NVIDIA, we could do that because our architecture is so proven. Number two, the scale of our customers is so incredible. Now, the scale of our supply chain is incredible, right? Who’s going to start all of that stuff, prebuild all of that stuff for a company unless they know that NVIDIA can deliver through? They believe that we can deliver through to all of the customers around the world.

They’re willing to start several hundred billion dollars at a time. The scale is incredible.

To that point, one of the biggest debates and controversies in the world is this question of GPUs versus ASICs: Google’s TPUs, Amazon’s Trainium, and it seems like everyone from Arm to OpenAI to Anthropic is rumored to be building one.

Last year, you said, “We’re building systems, not chips,” and you’re driving performance through every single part of that stack. You also said that many of these projects may never get to production scale. But given the seeming success of Google’s TPUs, how are you thinking about this evolving landscape today?

Yeah. First of all, the advantage that Google had is foresight. Remember, they started TPU 1 before everything started. This is no different than a startup. You’re supposed to build a startup—you’re supposed to create a startup—before the market grows. You’re not supposed to come up as a startup when the market is $1 trillion large.

This fallacy—and all VCs know this fallacy—that a large market, if you could just take a few percent market share, could make you a giant company is actually fundamentally wrong. You’re supposed to take 100% of a tiny market, a tiny industry, which is what NVIDIA did, right? Which is what TPUs did. There were only the 2 of us.

But you better hope that that industry gets really big. You’re creating an industry.

That’s right.

Right. And I mean, the NVIDIA story, you know—and so that’s the challenge for the people who are building ASICs now. It looks like a juicy market, but remember, this juicy market has evolved from a chip called a GPU to what I just described: an AI factory.

You guys just saw that I announced a chip called CPX for context processing and diffusion video generation—a very specialized workload, but an important workload inside a data center. I just alluded to maybe AI data-processing processors, because guess what? You need long-term memory. You need short-term memory. The KV-cache processing is really intense.

AI memory is a big deal. You’d like your AI to have good memory, and just dealing with all the KV caching around the system is really complicated stuff. Maybe it wants to have a specialized processor. Maybe there are other things, right?

You see, NVIDIA’s viewpoint is now not GPU. Our viewpoint is looking at the entire AI infrastructure and asking what it takes for these incredible companies to get all of their workload through it, which is diverse and changing.

Look at the transformer. The transformer architecture is changing incredibly. If not for the fact that CUDA is easy to operate on and iterate on, how do they try all of their vast number of experiments to decide which one of the transformer versions and what kind of attention algorithm to use? How do you disaggregate? CUDA helps you do all that because it’s so programmable.

The way to think about our business now is to look at when all of these ASIC companies or ASIC projects started, 3, 4, 5 years ago. I’ve got to tell you, that industry was super adorable and simple. There was a GPU involved, right? But now it’s giant and complex, and in another 2 years it’s going to be completely massive. The scale is going to be so large.

I think the battle of getting into a very large market as a nascent player is just hard, as you guys know—even for the customers who perhaps are successful with ASICs.

Isn’t there an optimal balance in their compute fleet? I think investors are very much binary creatures. They just want a yes-or-no, black-and-white answer. But even if you get the ASIC to work, isn’t there an optimal balance?

You think, “I’m buying the NVIDIA platform. CPX is going to come out for prefill, for video generation, maybe a decode platform—a video platform.”

Exactly.

Yeah. So there will be many different chips or parts to add to the NVIDIA ecosystem, the accelerated compute fleet, as new workloads are born.

That’s right.

And people trying to tape out new chips today aren’t really anticipating what’s happening a year from now. They’re just trying to get a chip to work.

That’s right.

Said another way, Google’s a big GPU customer.

Google’s a big GPU customer. Google is a very special case. We just have to show respect where respect is really deserved. TPU is on TPU 7, right? It’s a challenge for them as well.

There are 3 categories of chips. First, there are architectural chips: x86 CPUs, Arm CPUs, NVIDIA GPUs. They have an ecosystem above them, and the architecture has rich IP and a rich ecosystem. It’s very complicated technology. It’s built by the owners, like us.

There are ASICs. I worked for the original company, LSI Logic, that invented the idea of ASICs. As you know, LSI Logic is not here anymore, right? The reason for that is because ASICs are really fantastic.

When the market size is not very large, it’s easy to have somebody be a contractor to help you put the packaging of all that stuff together and do the manufacturing on your behalf, and they charge you 50%–60% gross margin. But when the market gets large for an ASIC, there’s a new way of doing things called COT: customer-owned tooling.

Who would do something like that? Apple’s smartphone chip—the volume is so large that they would never pay somebody else 50%–60% gross margin for an ASIC. They do customer-owned tooling. Where will TPUs go when they become a large business? Customer-owned tooling. There’s no question about it.

There’s a place for ASICs. Video transcoders will never be too large. Smart NICs will never be too large. When there are 10, 12, 15 ASIC projects going on at an ASIC company, I’m not surprised by that, because there are probably 5 smart NICs and 4 transcoders. Are they all AI chips? Of course not.

If somebody were to build an embedded inference processor for a specific recommender system and that was an ASIC, of course you could do that. But would you do that as the fundamental compute engine for AI that’s changing all the time? You’ve got low-latency workloads. You’ve got high-throughput workloads. You have token generation for chat. You have thinking workloads. You have AI video-generation workloads.

Now you’re talking about the workhorse backbone of your accelerated compute. That’s what NVIDIA is all about.

Again, dumb this down. It’s like playing chess and checkers, right? The fact of the matter is, the folks who are starting ASICs today, whether it’s Trainium or some of these other accelerators, et cetera, they’re building a chip that’s a component of a much larger machine.

You’ve built a very sophisticated system, platform, factory—whatever you want to call it—and now you’re opening up a little bit, right? You mentioned the CPX GPU, right? It seems to me that, in some ways, you’re disaggregating the workloads to the best slice of the hardware for that particular domain.

Well, we announced this thing called Dynamo, right? It’s disaggregated AI workload orchestration, and we open-sourced it because the future AI factory is disaggregated, right?

And you launched NVLink Fusion. That even says to your competitors, including Intel, which you just invested in, the way in which you participate in this factory that we’re building. Nobody else is crazy enough to try to build the entire factory, but you can plug into that if you have a product that’s good enough, compelling enough that the end user says, “Hey, we want to use this instead of an NVIDIA GPU, or we want to use this instead of your inference accelerator,” et cetera. Is that correct?

That’s right. We’re delighted to connect you in. Yeah.

Tell us a little bit about Fusion.

Such a great idea, and we’re so happy to partner with Intel on that. It takes the Intel ecosystem—you know, most of the world’s enterprise still runs on Intel—and the NVIDIA AI ecosystem, accelerated computing, and we fused it together, right? We did that with Arm, right? There are several others we’re going to be doing it with, and that opens up opportunities for both of us.

It’s a win for both of us, a great win. I’ll be a large customer of theirs, and they’re going to expose us to a much, much larger market opportunity.

Yeah. That’s deeply related to this idea. It’s the argument you’ve made that shocks some people: You say our competitors building ASICs have chips that are cheaper already today, but they could literally price all their chips at 0. They could price them at 0, and you would still buy an NVIDIA system because the total cost of operating that system—power, data center, land, et cetera—the intelligence output is still a better bet than buying a chip, even if it’s given to you for free.

Because the land, power, and shell are already $15 billion, right?

Yeah. We’ve taken a crack at the math on that. But walk us through your math, because I think for people who don’t spend as much time here, it just doesn’t compute. How could it possibly be that you were pricing your competitors’ chips at 0, given the expense of your chips, and it still is a better bet?

There are 2 ways to think about it. One way is, let’s just think about it from the perspective of revenues.

Yes.

Okay. So everybody’s power-limited, and let’s say you were able to secure 2 more gigawatts of power. Well, that 2 gigawatts of power, you would like to have translate to revenues.

Yes.

So your performance, or tokens per watt, was twice as high as somebody else’s tokens per watt because I did deep and extreme codesign, right? My performance was much higher per unit of energy. Then my customer can produce twice as much revenue from their data center. And who doesn’t want twice as much revenue?

If somebody gave them a 15% discount, the difference between our gross margins, which is, call it, 75 points, and somebody else’s gross margins, call it 50 to 65 points, is not so much as to make up for the 30× difference between Blackwell and Hopper. Let’s pretend Hopper is an amazing chip, an amazing system. Let’s pretend somebody else’s ASIC is Hopper. Blackwell is 30×.

So you’ve got to give up 30× revenues in that 1 gigawatt. It’s too much to give up. Even if they gave it to you for free, you only have 2 gigawatts to work with. Your opportunity cost is so insanely high. You would always choose the best performance per watt.

I heard this from one of the CFOs at one of the hyperscalers: Given the performance improvement that’s coming out of your chips, again, precisely to that point—tokens per gigawatt, and power being the limiting factor—they had to upgrade to the new cycle. When you look ahead at Rubin, Rubin Ultra, and Feynman, does that trajectory continue? We’re building 6 or 7 chips a year now?

Yeah, and each one is part of that system.

That’s right.

And that system software is everywhere, and it takes the integration and the optimization across all of those 6 or 7 chips to deliver on the 30× Blackwell. Now imagine I’m doing this every single year. Bam, bam, bam, bam, bam, bam. And so if you build 1 ASIC in that soup of ASICs, in that soup of chips, and we’re optimizing across that, you know, it’s a hard problem to solve.

8. Nvidia's Competitive Moat

This does bring me back to where we started, about the competitive moat. We’ve been covering this as investors for a while. We’re investors throughout the ecosystem and in competitors of yours, from Google to Broadcom. But when I really just go to first principles around this and say, are you increasing or decreasing your competitive moat? You moved to an annual cadence. You’re co-developing with a supply chain. The scale is massively bigger than anybody anticipated, which requires scale both of balance sheet and of development.

The moves you made, both through acquisition and organically, with things like NVLink Fusion and CPX, which we just talked about—all of those things together cause me to believe that your competitive moat is increasing vis-à-vis, at least insofar as building out the factory or the system. It’s at least surprising.

But I think it’s interesting that your multiple is much lower than most of those other people. I think part of that has to do with this law of large numbers: A $4.5 trillion company couldn’t possibly get any bigger. But I asked you this a year and a half ago. As you sit here today, if AI workloads are going to 10× or 5×, and we know what capex is doing, et cetera, is there any conceivable world in your mind where your top line in 5 years isn’t 2 or 3× bigger than it is in 2025? What’s the probability that it’s actually not much higher than it is today, given those advantages?

I’ll answer it this way: Our opportunity, as I described it, is much larger than the consensus. I’ll say it here: I think NVIDIA will likely be the first $10 trillion company.

I’ve been here long enough. It wasn’t that long ago, just a decade ago, as you well remember, that people said there could never be a trillion-dollar company. Now we have 10, right? And today the world’s bigger, right? This is back to the exponentials around GDP and the growth. The world is bigger.

People misunderstand what we do. They remember we’re a chip company, right? And we are—we build chips. Boy, do we build chips, and build the most amazing chips in the world. But NVIDIA is really an AI infrastructure company. We are your AI infrastructure partner, and our partnership with OpenAI is a perfect demonstration of that.

Yeah.

We are their AI infrastructure partner, and we work with people in a lot of different ways. We don’t require anybody to buy everything from us. We don’t require that they buy the full rack. They could buy a chip. They could buy a component. They could buy our networking. We have customers buying only our CPU.

Just buy our GPUs and buy somebody else’s CPUs and somebody else’s networking. We’re okay selling any way you like to buy. My only request is, just buy a little something from us.

You said this isn’t just about better models. We also have to build. We have to have world-class builders. And you said the most world-class builder maybe that we have in the country is Elon Musk.

We talked about Colossus 1 and what he was doing there, standing up a couple hundred thousand, at the time, H100s and H200s in a coherent cluster. Now he’s working on Colossus 2, which may be 500,000 GPUs—millions of H100 equivalents—in a coherent cluster. I wouldn’t be surprised if he gets to a gigawatt before anybody else does in AI.

Say a little bit about that—the advantage of being the builder who isn’t just building the software and the models, but understands what it takes to build those clusters.

Well, these AI supercomputers are complicated things. The technology is complicated. Procuring it is complicated because of financing issues. Securing the land, power, and shell, powering it, is complicated. Building it all and bringing it all up—I mean, these are, unfortunately, the most complex systems problems humanity has ever endeavored to solve.

Elon has a great advantage that, in his head, all of these systems are interoperating, and the interdependencies reside in one head, including the financing.

Yes. He’s a big GPT. He’s a big supercomputer himself.

He’s the ultimate GPU. And so he has a great advantage there. He has a great sense of urgency. He has a real desire to build it, and so when will comes together with skill, unbelievable things can happen.

Yes. Yeah. Quite unique.

9. Sovereign AI & Global Buildout

Something you’ve been so involved in is—I want to talk about sovereign AI. I want to talk about China and the global AI race that’s going on. When I look back at you 30 years ago, you couldn’t have imagined you were going to be hanging out in palaces with emirs and the king this week, and you’re at the White House all the time.

The president has said that you and NVIDIA are critical to U.S. national security. So, when you look at that, first contextualize it for me. It's hard to believe that you would be in those places if sovereigns didn't view this as at least existential—as important as maybe we viewed nuclear weapons in the 1940s, right? We don't have a Manhattan Project today, at least one funded by the government, but it's funded by NVIDIA, OpenAI, Meta, and Google. We have companies today the size of nation-states funding something that it appears to me presidents and kings think is existential to their future economic and national security. Would you agree with that?

Nobody needs atomic bombs. Everybody needs AI.

Bill Gurley

Well said. Here, here. Yeah. Here.

Okay. And so that's a very, very large difference. AI, as you know, is modern software. That's where I started from: general-purpose computing to accelerated computing, from human-written code one line at a time to AI-written code. That foundation can't be forgotten. We've reinvented computing. There's not a new species on Earth; we just reinvented computing. Everybody needs computing, and it needs to be democratized.

Which is the reason why all of these countries realize they have to get into the AI world, because everybody needs to stay in computing. There's nobody in the world that says, "Guess what? I used to use computers yesterday. I'm pretty good with clubs and fire tomorrow." Everybody needs to move into computing. It's just being modernized, that's all.

Number 1, it is the case that in order to participate in AI, you have to encode within AI your history, your culture, and your values. Of course, AI is getting smarter and smarter, so even the core AI is able to learn these things fairly quickly. You don't have to start from ground zero. I think that every country needs to have some sovereign capability.

I recommend that they all use OpenAI, Gemini, these open models, Grok, and Anthropic. But they should also dedicate resources to learn how to build AI. The reason for that is because they need to learn how to build it not just for language models, but for industrial models, manufacturing models, and national security models. There's a whole bunch of intelligence they have to cultivate themselves. They ought to have sovereign capability. Every country should develop it.

Is that what you see? Is that what you're hearing around the world?

10. The AI Administration

They all realize it. They all are going to be customers of OpenAI, Anthropic, Grok, and Gemini, but they all really need to also build their own infrastructure. This is the big idea: what NVIDIA does is build infrastructure. Just as every country needs energy infrastructure and communications and internet infrastructure, now every single country needs AI infrastructure.

So, let's start with the rest of the world. You know, our good friend David Sacks. The AI czars are doing a heck of a job.

Bill Gurley

We are so lucky.

Yeah, to have David and Sriram in Washington, D.C. David doing AI as the AI czar—what a smart move by President Trump to put them in the White House. Because during this pivotal time, the technology is complicated.

Bill Gurley

Yes.

Sriram is the only person in Washington, D.C., who I think knows CUDA.

Bill Gurley

Yeah.

And that's strange, anyways. But I just love the fact that during this pivotal time, when technology is complicated and policy is complicated, the impact on the future of our nation is so great that we have somebody who is clear-minded, dedicating the time to understand the technology and thoughtful enough to help us through that.

It would seem to me—I'm going back to the Manhattan Project analogy—that you have a president who understands how existential this is. You have governors like Greg Abbott in Texas who want to remove regulations to accelerate because they understand how important it is. You have Secretary Wright at Energy, Doug Burgum at Interior, and Lutnick at Commerce, who also understand how important this is and how pro-energy they are. Could you imagine the alternative if we had an administration right now that was not pro-energy and didn't want energy to grow in our nation so that we could have AI?

I just can't even think about it. I find it ironic that just a couple of years ago we were saying, "China's building 100 nuclear reactors. They're so far ahead of us." That's the prerequisite to AI. But now, when we go to build it, everybody says, "Oh, it's a glut," right?

It seems to me that this is something that the government—it is in their interest. We have industry and government working together in a way that I haven't seen in a long time. You've been around a long time. You're very close with President Trump at this stage. Help us understand: what is the nature of industry-government relationships? We saw that dinner last week with all the CEOs. You spent a lot of time there. Is it unique? Have you seen anything like this in your career over the last 30 years?

It was hard to go to D.C. in the past, as you know. Getting an appointment was almost impossible.

Bill Gurley

Right?

President Trump has an open door to leaders who want to come in and help him understand the future. This is an administration that believes in growth. Fundamentally, President Trump wants America to grow.

Bill Gurley

Yeah.

If we can grow economically, we will be strong militarily. If we could grow economically, we will be secure. I've never met somebody who is secure who's poor. Being rich as a nation is an essential part of national security, and he knows that.

He also wants America to win the AI race. This is going to be a very long-term race, and he understands that this is a pivotal time. He wants the technology industry to run. He wants everybody in the world to be built on American technology.

These are sensible, logical things. The opposite is strange to me. If I take everything and just reverse it, we want our country not to grow. Because we don't want our country to grow, we don't need any energy, because we know we need energy to grow, so let's not have any energy. In fact, we don't want our technology industry to lead. He understands that our technology industry is our national treasure.

Bill Gurley

Correct.

Technology, like corn and steel and things in the past, is now such a fundamental trade opportunity. It's an essential part of trade. Why would you not want American technology to be coveted by everyone so that it could be used for trade?

Right. So let's talk about the internet.

Yeah.

Google spread around the world. We had democratic values spread around the world by way of search, and Google didn't have to go to Washington to get permission to do it. It just happened. We diffused our technology around the world. David Sacks has been crystal clear about the need to accelerate export licenses so that the American AI stack wins around the world. We're talking chips, models, data centers, et cetera.

We know a year and a half ago that wasn't happening. There was a concept called "small yard, high fence" or something like that. A small yard, high fence. The irony of it was it was described and recommended in policy in such a way that it was a small yard, high fence around America. That was the strange part. I think President Trump has got it right: we want to maximize exports. We want to maximize American influence around the world. We're supposed to maximize those things.

Do you see those licenses coming? Are you seeing the acceleration in Washington? I know it's being said at the top, but are you seeing it flow down through government that's accelerating us around the world?

Secretary Lutnick was all over it.

Bill Gurley

Great. Yeah.

11. Chinese AI Chips & NVIDIA’s Role

So now let's talk about China. You know what most people may not realize is, I think you understand China as well as any leader in the United States.

We've been there for 30 years.

Been there for 30 years. What most people don't realize is that, up until a couple of years ago, you had dominant market share within China in terms of—

95% market share.

95% market share in, arguably, the most important thing. And you have said that our biggest own goal, as a country, under the guise of somehow trying to slow them down, is that we've unilaterally disarmed. We forced NVIDIA out of China, which has allowed Huawei to accelerate on the back of monopoly profits within China.

I just saw this morning announcements out of Huawei, Alibaba, and others that they're going to build data centers around the world. Now Huawei has a 3-year plan to pass NVIDIA, funded by the monopoly profits in the biggest AI market in the world. So it's looking like your admonition that this is a huge mistake—to hand China monopoly markets—is coming true.

The president said, after the ban on H20s, that now we have a situation where you can sell chips to China, but there's a 15% export tax.

But now it appears that the Chinese, perhaps offended by statements out of the United States, are saying, no, NVIDIA is not allowed to sell here. Where do we stand today between NVIDIA and China? And can you reiterate what you think we as a country should be doing to put ourselves in the best position to win the AI race around the world?

We have a competitive relationship with China. We should acknowledge that China rightfully should want their companies to do well. I don't, for a second, begrudge them for that. They should do well. They should give them as much support as they like. It's all their prerogative.

And don't forget that China has some of the best entrepreneurs in the world because they came from some of the best STEM schools in the world. They're the most hungry in the world.

Yes.

996, as you know, producing the most AI engineers in the world.

996. So the audience knows: 9 in the morning to 9 at night, 6 days a week.

That is their culture.

Yeah.

Okay. We're up against a formidable, innovative, hungry, fast-moving, underregulated—

Yeah.

Okay. People don't realize this. They are very lightly regulated.

Right? Less regulated, ironically, than we are in a capitalist system.

That's right. People think that they're centrally governed. But remember, the genius of China was distributed economic systems. All of these 33 provinces and all the mayoral economies have driven an enormous amount of internal competition and internal economic vibrancy, which, of course, has some of its side effects.

But this is a vibrant, entrepreneurial, high-tech, modern industry. Some of the things I heard were, one, they could never build AI chips. That just sounded insane. Two, that China can't manufacture. China can't manufacture? If there's one thing they could do, it's manufacture. And three, they're years behind us. Is it 2 years, 3 years? Come on. They're nanoseconds behind us.

Nanoseconds.

Yeah, they're nanoseconds behind us. And so we've got to go compete.

Yeah.

We've got to go compete.

And so the question then becomes: What's in the best interest of China? Of course, it's that they have a vibrant industry. They also publicly say—and rightfully, I believe they believe this—that they want China to be an open market. They want to attract foreign investment. They want companies to come to China and compete in the marketplace.

Right?

And I believe—and I hope—that they would return to that. In our context, answering your question, what do I see in the future? I do hope, because they say it, their leaders say it, and I take it at face value—and I believe it because I think it makes sense for China—that what's in the best interest of China is for foreign companies to invest in China, compete in China, and for them to also have vibrant competition themselves. They would also like to come out of China and participate around the world.

That, I think, is a fairly sensible outcome. What we need to do as a country is enable our technology industry. I'm privileged to be working in an industry that is our national treasure. We have to acknowledge it is our national treasure. It is our best industry.

It is our single best industry.

Yeah. Why would we not allow this industry to go compete for its survival? Why would we not allow this industry to go and proliferate the technology around the world so that we could have the world be built on top of American technology? That way, we can maximize our economic success, maximize our geopolitical influence, and maximize this technology industry during such a vibrant, pivotal time. We should allow it to thrive.

The skeptic says Jensen just wants to sell more chips, and if he can sell them to China, great, he'll sell them to China. He doesn't care about what that means for America. That's the skeptic. Now—

Can I just address the skeptics? Just because I want America's ecosystem and economy to grow doesn't make me wrong. Okay? First of all, everything that's been said so far, that's been made up about U.S.-China, has proven to be wrong. The facts are just wrong. The ground truth is wrong.

Just because we want America to win, just because we want this industry to grow, doesn't make me wrong.

Correct. And I think anybody who knows you, and now the president, certainly myself, knows that you deeply care about the country. You deeply want the United States of America to win the global AI race. You just happen to believe—and I think you have as much or more experience than anyone—that it inures to our advantage. The probability of us winning the global AI race actually goes up if you are competing in China because it allows us to tap into half of the world's AI engineers, keeping them in this ecosystem.

Let's be clear: The companies we're talking about here—ByteDance, Alibaba, et cetera—are companies that are largely owned by American investors.

Yeah. Right. These are global companies that are building recommendation engines that, by the way, are extraordinary technologies, incredible companies. And so I think—and I'm hopeful—that the argument that you're making vis-à-vis China, which is a harder argument than diffusion to the rest of the world—I understand that. That's why I thought when the president said, “I don't know, it's a flip of a coin. Maybe Jensen's right. Maybe the other guys are right. If Jensen's willing to put a little bit—15%—into the U.S. Treasury as a hedge on that, then I'll go for it.”

But I was really disappointed on the heels of that.

Mhm.

I think if the Chinese feel like they're being taken advantage of, that we're going to send them chips that are 10 years old or something, then I get why they had that response.

H20 is really quite spectacular still. Of course, it's not as good as Blackwell, and I get that.

Yeah.

Look, I'm patient, and I believe that they're wise. They're thinking through their situation. They have larger agendas to deal with. There are a lot of discussions going on in the United States.

But I'll come back to the ground truth, the fundamental truth. I believe it is in the best interest of China that NVIDIA is able to serve that market and compete in that market. I fundamentally believe it is in the best interest of China. It is, of course, in the fantastic interest of the United States.

Yeah.

It is fantastic. But those 2 truths can coexist. It is possible for both to be true, and I believe they are both true.

And so, even though I tell all of our investors that our guidance includes no China—

Yeah.

—and I appreciate all of our investors including no China in any of our guidance, we've got plenty of growth opportunities outside, and all of that is true. It doesn't make China not important to us. It's very important to us. Anybody who thinks that the Chinese market is not important has their head deep in the sand.

Yeah.

This is one of the most important markets in the world. Smart markets, as you know—smart people doing smart things—and we want to be there. I think it's in the best interest of both countries that we are there.

When I take a step back, I am confident that ultimately wisdom will prevail.

Yes.

I've always been confident that wisdom prevails. I've always been confident that truth prevails, and it's taken me this far. I believe that to be fundamentally true now. These things will get sorted out, and we will have the opportunity to go compete in that China market.

12. H-1B, Talent, & the American Dream

I'm not very political, but this is very topical: the administration's decision to charge $100,000 per H-1B visa. You've spent a lot of time with the president. You've called him our secret weapon in AI. I also know you want to recruit the best and brightest to our country.

So how do you think about the decision to charge $100,000 per H-1B visa? Does this make it easier or harder to recruit talent? And perhaps it's a little different for large companies or small companies. How do you think about it?

I'm going to start with: it's a great start.

Hold on. You said it's a great start.

It's a great start. I'm just going to start there, and the reason for that is this.

That implies you hope it's not the end.

I hope it's not the end, but I think it's a great start. I just hope it's not the end.

Here's what I fundamentally believe: America has a singular brand reputation that no country in the world has. No country in the world is in a position, or on the horizon, to be able to say, “Come to America and realize the American dream.”

What country has the word “dream” behind it?

Yes, it's part of its brand. We are utterly singular, and you're talking to somebody who represents the American dream. My parents didn't have any money. They sent us over here. We started from nothing. You guys know I busted tables, washed dishes, cleaned toilets, and here I am.

Yeah.

This is the American dream. President Trump knows that we want legal immigrants.

Yeah.

There's a difference between legal immigrants and illegal immigrants. But the idea that it's a country that's free-for-all doesn't make sense.

And so now the question is: How do we go from the idea that we want to protect, fundamentally, the American dream to dealing with illegal immigrants at such a large scale? How do we find a logical, pragmatic solution?

So the idea that we put a $100,000 price tag on H-1B probably sets the bar a little too high, but as a first bar, it at least eliminates illegal immigration, and that's a good start.

How does it eliminate illegal immigration?

Well, it at least eliminates abuse of H-1B. Yeah, at least. And that's a good start, and at least we can have a conversation.

So, one of the things that we know about President Trump is that he's a good listener. He actually listens. He listens to you, he listens to me, and he doesn't have to. He listens to a lot of people, and he's integrating a lot of information. This is obviously a very complicated issue.

And so I think that this is a fine start. It's a fine start. But I'm not confused that anyone in the administration, anyone in the White House, is confused about the fact that legal immigration is the foundation of the American dream and is the ultimate brand that we want to protect. That's the future we want to protect.

And I would also say it seems to me that certainly David Sacks and other people in the administration know that we have to recruit the world's best and brightest. We should not sacrifice the greatness of the brand. Charging $100,000—or, let's say, it got lowered to $50,000 or whatever the case is—it does seem like it tilts the playing field in favor of big companies that can effectively sponsor all these people, right? And it's more challenging for the startup ecosystem, where people are already super expensive, and now I have to pay this fee on top of it.

It also has an unintended consequence. It might accelerate investment outside the United States, right? And so there are unintended consequences, but like I said, start somewhere, move toward the right answer, right? Oftentimes, people want to go directly from a wrong answer, a wrong condition—we don't want this condition where we're at, right?—and directly jump to the perfect answer. The perfect answer is hard to find, right? Just start somewhere. It's the entrepreneurial way.

It's important to me. The president talked before, when he was running for office, about wanting to staple a green card to the diplomas of these STEM students. Smart people coming to the United States from China, AI researchers studying at Stanford—we want to keep them here. And by the way, if their families can't get here, they're going to leave after a few years, so you might even want to make it easier for their families to come here.

Are you confident that we have a strategic plan in this administration? This is a start, but your conversations give you confidence that we have a broader strategic plan to make sure we're recruiting the best and the brightest?

I don't know that I have an answer for that.

Okay.

But I understand that where we're at is not where we want to be. And I don't think anybody's lost their focus on the American dream, the importance of immigration, the importance of attracting all of the world's best talent to the United States, and creating the conditions for them to stay here. There are things that are done from time to time that work against what I just described, right? Making foreign students uncomfortable and being here threatens the brand. Let's not forget that it's okay to be competitive with China, but be careful not to be tough on Chinese. And so we need to make sure that that slippery slope isn't crossed.

Yeah.

You know, there are all of these things that go along with finesse and nuance. But the fact of the matter is we know where we want to be. We know we're in a difficult situation. We don't want to be here, and President Trump doesn't have much time to move us in that direction.

Right.

And so, to the extent that we move in that direction, I believe it's a good start.

13. Invest America & American Right to Rise

Agreed. Yeah, I heard from a Chinese researcher leading one of our leading labs in the U.S. that 3 years ago, 90% of the top AI researchers graduating from universities in China wanted to come to the United States and did come to the United States to work in our leading labs, and he guessed that today that's closer to 10% or 15%. Right? So we've seen a precipitous drop. That's precisely a concern that we have, right?

So have you seen this? You're paying attention to both markets. Do you see this, and what are the things we need to do in order to reverse that?

I definitely see a greater concern among Chinese students who come here and remain here. And many of them who come here for school are thinking about going elsewhere, right? Many of them are thinking about Europe, right? And so I think we need to be super, super concerned about this. This is a source of existential crisis. These are definitely early indicators of a future problem.

Right, right. Smart people's desire to come to America and smart students' desire to stay—those are what I would call KPIs.

Yes.

Early indicators of future success.

Yes.

I think of it a bit like the Warriors. If they have an advantage in recruiting all the best players in the NBA, right, then they can continue to win championships. But the second that recruiting pipeline, because of the brand of the Warriors, gets diminished or something else happens, then they're not going to be able to recruit the best future players, and they're not going to win championships.

And when you talk about the American dream so eloquently, that being Brand USA—the right to come here and to do what you've done—I hope that the feedback to this administration isn't just about the administration; it's also about how we as a country talk about immigration.

That's right.

Right. This needs to be the place that welcomes the best and the brightest, that attracts them, has a strategic plan for recruiting the best and the brightest, and makes sure that this is the place where they want to work.

As you know, there's a phrase—and I didn't hear about this phrase until just a few years ago—China hawks.

Yes.

And apparently, if you're a China hawk, you get to wear that label with pride. It's almost like a badge of honor, right?

It's a badge of shame. There's no question it's a badge of shame. There's no question that although they want what's in the best interest of our country, and we all want what's in the best interest of our country—

Right.

—destroying that pipeline of the American dream—

Yeah.

—is not patriotic, right? They think they're doing the right thing for our country, but it's not patriotic. Not even a little bit. And so we need to continue to be the great country we are, to have the confidence of a great country.

Yes. Well said.

And to have the confidence of a great country and have somebody who wants to compete with us, and to have the attitude: bring it on.

Right, right.

Bring it on.

Right. Because I believe in our people. I believe in the people that are here. I believe in our culture. I believe in our country. I believe in our system. Bring it on.

And is it your take that that's where the president is? He's a pragmatist. He's a believer in the growth and the ability of the United States to compete. It seems to me that's where he is.

There's no question President Trump is the bring-it-on president.

Right, right. And he doesn't seem to me like—the reason I'm confident, and I've said on this pod that I think he'll get a big deal done with China—

I really, really do hope so.

Yeah. And I think he speaks positively, with great respect and great eloquence, about his relationship and the importance of China. Not 1 time have I ever heard him say the word “decouple,” which we heard a lot in the last administration, right? You can't decouple the 2 most important relationships for the next century. That doesn't make any sense at all. Decoupling is exactly the wrong concept.

I mean, it seems to me he and Scott Bessent are saying, “Listen, we need to make America great. We need to reindustrialize America. We need to balance and make sure that we have fair trade, that we protect industries that we need to help build, that China helps us do that, recognizing that we have helped them do it over the course of the last 25 years.”

But ultimately, he said, “The best way to understand me is I'm a great dealmaker. I make deals,” right? Whereas I think in other camps there are people who are iconoclastic or dogmatic. It's the Mearsheimer view of China, that there's a great-power struggle: one must win and one must lose, versus the idea that every country has to look exactly like ours, right? You know, we want diversity.

You want America to win, but that doesn't have to come at the expense of poking somebody in the eye and telling somebody else they have to lose, because we're that confident.

Yeah, we're that confident.

Because we're that mighty. Because we're that incredible. I've got no trouble, as you know, working with all my colleagues in the ecosystem, right? And notice we just did the ultimate deal, right? Partnering with Intel, a company that spent most of its life trying to put us out of business, right? And I had no trouble partnering with them, right? You know, and the reason for that is because, number 1, bring it on.

Yes. And number two, the future is so much greater. It doesn't have to be all us or them. It could be us and them. But nonetheless, bring it on.

14. Elon, X.ai & Colossus 2

Yeah, agreed. You mentioned something that's profoundly important to both of us. You and I have talked a lot about the American dream, and it was, I think, Abraham Lincoln who said, “Fundamental to the American dream is the right to rise.”

Yeah, that's right.

The belief that your kids can do better than you did.

That's right.

Right. And you've experienced the right to rise. We've all experienced the right to rise in America.

So, yeah, you go to Wikipedia, you look up “American dream,” and there's my picture, right? The ultimate American dream.

And yet we live at this time where, because of the nature of these technological systems, we have companies that are going to be worth $10 trillion. We'll probably have individuals who are worth $1 trillion. Those are the incentives that give people the encouragement to rise.

At the same time, when we head into this age of abundance, something that I was deeply worried about was that too many people get left behind. They feel left out and left behind, so it makes sense for them to attack this system of capitalism.

Something that you and I worked on together, and I'm deeply grateful for, was the idea of Invest America: that we have to start every kid at birth on the capitalist right-to-rise journey. Give them $1,000 in great companies like NVIDIA, SpaceX, and OpenAI, et cetera. They benefit as the country wins; they win, and they own it individually. Every kid is a shareholder in the future of America.

Well, I want to thank you for starting it, for driving it.

Yeah.

Yeah, what a great idea.

And, you know, so this—

You're a genius.

The thing is, this passed in the One Big Beautiful Bill. Most people don't even realize that yet. Starting in 2026, every child born forevermore in the history of this country will start off with an investment account at birth, seeded with $1,000 in the best American companies.

Your company has agreed to add to the accounts not only of the kids of your employees but maybe even kids of others. I'm going to adopt schools, you know, and lots of philanthropists and companies. We think every company across America should do this. It's a wonderful way for companies to give back, right?

Yeah, as part of the 401(k).

This seems to me to be part of the change in the social contract that needs to occur, because if we're seeing this exponential progress, we know that the evolution of government and the social contract needs to keep up with it.

Obviously, President Trump and a bipartisan group in the House and Senate passed this into law. So maybe just talk to us a little bit about how you think about the pace and magnitude of the changes that are coming. I know you believe it will be a net good, but there are also going to be a bunch of people displaced along the way. We probably need things like this, and other things, in order to bring everybody along for the journey.

There are several things that President Trump has done that are incredibly good for bringing everybody along. Let me just start there. The first thing is reindustrializing America.

President Trump, Secretary Lutnick—they're all behind that. All the work that they're doing, encouraging companies to come build here in the United States, investing in factories, and reskilling and upskilling that skilled labor workforce is incredibly valuable to our country.

The idea that we no longer make it so that you have to get a PhD or go to one of the great schools, and only in that way can you build a great life and deserve to have a great living—we've got to change all that. It doesn't make any sense. We love craft. I love people who make things with their hands, and now we're going to go back and build things.

And build magnificent, incredible things. I love that.

That's going to transform America. There's no question about that. There's a whole band of the economy, a whole band of society, that has been largely left behind because we outsourced everything, right?

I'm not suggesting we insource everything. All the people arguing about manufacturing tennis shoes and toothpicks—I mean, that's denigrating a perfectly good discussion into some insane level. We've got to recognize that reindustrializing America is just fundamentally going to be transformative, number one.

And aspirational.

Oh, it's fantastic.

Elon taking us to Mars, watching spaceships caught with chopsticks out of the sky—this is not only great for the industrial base of America, it's aspirational.

Fantastic. That's right.

And then, of course, AI.

Yeah, it is the greatest equalizer. Just think: everybody can have an AI now. The ultimate equalizer. We've closed the technology divide.

Remember the last time that somebody who wanted to use a computer for their economic or career benefit had to learn C++ or C, or at least Python. Now they just have to learn human language, you know. If you don't know how to program an AI, you tell the AI, “Hi, I don't know how to program an AI. How do I program an AI?” And the AI explains it to you—

Or does it for you.

It does it for you. And so it's incredible, isn't it? We've now closed the technology divide with technology.

Yeah.

This is something that everybody's got to engage with. OpenAI has 800 million active users. Gosh, it really needs to be 6 billion.

Yeah.

Right. It really needs to be 8 billion soon. And so I think that's number one. Then number two—and then number three—I think AI will change tasks.

The thing that people confuse is that there are many tasks that will be eliminated. There are many tasks that will actually be created. But it is very likely that, for many people, their jobs are gainfully protected.

For example, I'm using AI all the time. You're using AI all the time. My analysts are using AI all the time. My engineers—every one of them uses AI continuously. And we're hiring more engineers. We're hiring more people. We're hiring across the board.

The reason is that we have more ideas. We can now go pursue more ideas. The reason for that is because our company became more productive. Because we became more productive, we became richer. Because we became richer, we can hire more people to go after those ideas, right?

The concept that AI comes along and therefore there's going to be a mass destruction of jobs starts with the premise that we have no more ideas.

Right.

It starts with the premise that we have nothing left to do. Everything we're doing in our lives today—this is the end.

Yeah.

And if somebody else were to do that one task for me, I have one task left. Now I have to sit there and wait for something.

Yes.

You know, wait for retirement, sit on my rocking chair. That idea doesn't make sense to me.

I think intelligence is not a zero-sum game. The more intelligent people I'm surrounded by, the more geniuses I'm surrounded by, surprisingly, the more ideas I have, the more problems I imagine that we can go solve, the more work we create, and the more jobs we create.

I don't know what the world looks like in a million years. That's going to be left for my children. But for the next several decades, my sense is that the economy is going to grow. Lots of new jobs are going to be created. Every job will be changed. Some jobs will be lost.

15. The Future Ahead

We're not going to be riding horses on streets and those things. It'll be fine. Humans are famously skeptical and terrible at understanding compounding systems, and they're even worse at understanding exponential systems that accelerate with size.

We've talked about exponentials a lot today. The great futurist Ray Kurzweil said that in the 21st century, we're not going to have 100 years of progress. We're likely to have 20,000 years of progress.

Right.

You said earlier we're so fortunate to be living at this moment and contributing to this moment. I'm not going to ask you to look out 10 or 20 or 30 years, because I think it's so challenging. But when we think about things like robots—

30 years is easier than 2030.

Oh, really?

Yeah, yeah.

Okay, so I'll grant you license to go out 30. As you think out over the course of 30 years—

I like these shorter time frames because they have to marry bits and atoms. Bits and atoms are the hard part of building this stuff, right?

Because everybody saying it's going to happen is interesting but not helpful.

Exactly.

But if we have 20,000 years of progress, reflect on that statement by Ray, reflect on exponentials, and think about how all of our listeners—whether you work in government, whether you're in a startup, whether you're running a big company—need to be thinking about the accelerating rate of change, the accelerating rate of growth, and how you will be co-intelligent in this new world.

Well, there are a lot of things that many people have already said, and they're all very sensible. I think in the next 5 years, one of the things that's really cool that's going to get solved is the fusion of artificial intelligence and mechatronics—robotics. And so we're going to have AIs that are going to be wandering around us.

And we all know that we're all going to grow up with our own R2-D2.

And that R2-D2 will remember everything about us, coach us along the way, and be our companion. We already know that.

Yeah.

Okay. And so the idea that every human will have their own GPUs associated with them in the cloud—and that there are 8 billion people, 8 billion GPUs—that's a viable outcome.

Yeah, you know, and so—and each having their own model that's fine-tuned for them.

Fine-tuned for them. And that AI that's in the cloud is also embodied in your car. It's embodied in your own robot. It's everywhere with you.

And so I think that future is a very sensible thing. The idea that we're going to understand the infinite complexity of biology, understand the system of biology, how to predict it, and have digital twins of everybody. Our own digital twin for health care, like we have a digital twin for shopping at Amazon. Why wouldn't we have our digital twin in health care? Of course we would.

And so, a system that predicts how we're going to age, what disease we're likely to have, and anything that's about to happen—maybe even next week or tomorrow afternoon—and predicts it early. Of course, we're going to have all that.

I think the part that I'm asked a lot by CEOs that I work with now is, given all of that, what happens? What do you do? And this is the common sense of things that move fast.

Right?

If you have a train that's about to get faster and faster and go exponential, the only thing that you really need to do is get on it.

Yeah.

And once you get on it, you'll figure everything else out along the way.

Right.

And so, to predict where that train's going to be—

Right.

—and try to shoot a bullet at it, or predict where that train's going to be when it's going exponentially faster every second and go figure out what intersection to wait for it—

Right?

That's impossible. Just get on it while it's going kind of slowly, and go exponential along the way.

A lot of people think this just happened overnight. You've been at this for 35 years. I remember hearing Larry Page say, probably around 2005 or 2006, that the end state of Google would be when the machine could predict the question before you even ask it and give you the answer without having to look.

Because, contextually, you must be asking about—you must be wondering about that.

I heard Bill Gates say in 2016, when somebody said, “Haven't all the things been done? We've had the internet, we've had cloud, we've had mobile, social, et cetera.” He said, “We haven't even started.” They said, “What do you think? Why would you say that?” He said, “We won't even begin until machines go from being dumb calculators to beginning to think for themselves, to think with us.”

That's the moment that we're in. I think to have leaders like you, leaders like Sam and Elon, Satya, et cetera, it's such an extraordinary advantage for this country.

And to have the cooperation that we see between a system of risk capital that I'm part of, which can provide the risk capital for people to do this—we're not relying on the government to have a Manhattan Project. We can actually do this ourselves, together, for the benefit of the country. It's an extraordinary time.

And at a scale that's unimaginable.

Right. Right. It's an extraordinary time. But I also think one of the things that I'm just grateful for is that we have leaders who also understand their responsibility to the fact that we are creating change at an accelerating rate. And we know that while it will most likely be great for the vast majority, there'll be challenges along the way. And we'll deal with those as they come.

And raise the floor for everybody and make sure that this is a win, not just for some elite bureaucrats at the top hanging out in Silicon Valley. And don't scare them. Bring them along. Don't scare them. Bring them along.

And we will.

Yeah.

So, thank you for that.

Exactly.