Jensen Huang：唱衰 AI 的人看错了

1. The fight over Chinese AI

Axios always starts with the news. The Financial Times says Chinese regulators are looking at export controls, tightening their controls on AI models and semiconductors to keep them away from the West. Do you think that will happen, and what would the net effect on the U.S. be?

I hope not. The Chinese market and the Chinese industry create some really great AI models and AI technology. I've said before that half of the world's AI researchers are Chinese. Many of them still live in China, and they produce some really groundbreaking research in AI.

The United States does, too, and many places around the world do. We want to make sure that we can all share ideas as broadly and as quickly as possible so that we can advance this incredible technology in a way that's beneficial to society and safe. I hope that they continue to keep it open.

You were just in China a couple of months ago. Do you think they will? What's the likelihood?

I don't know. I don't know the source of that news. Of course, all kinds of regulation are being discussed, and export controls are being discussed on both sides of the ocean.

Once one side decides to exercise export controls, everybody decides to think about export controls. I think we all have to de-escalate and let the technology advance. The companies in the United States are very, very capable. We're able to move very quickly.

I have every confidence that the United States will continue to lead. So long as we have the support and the encouragement of governments, we should be able to continue to do so.

2. Should the U.S. restrict Chinese AI models?

Flip side: Should the government ban or restrict Kimi or other Chinese models?

I hope not, and I really don't think so. The United States and the world need open and closed models. We should use closed models because they're great to use. They work incredibly well. Anthropic and OpenAI—these are incredibly great services. We should use closed models as much as we can and rent everything we can.

However, the industry still needs open models. Science needs open models. Cybersecurity needs open models because they're safe. It's important for cybersecurity, national security, and economic security. Companies need to be able to control some of the IP and some of the data that they have so that they can advance the AI models in a controlled way.

Cybersecurity, with the benefit of transparency and the ability for the entire community to inspect and test to make sure that we harden it in the best possible way, is best served by open models. And so, irrespective of how many industries use them, open models are foundational to the advance of science and the advancement of computer science today. I really do hope that we keep it open.

3. Should American companies use Chinese AI?

A simple question on the front page of The Wall Street Journal: Should American companies be allowed to use Chinese AI models?

Absolutely. There's a misconception that somehow there are back doors, that they're somehow connected to China in some way. You download the models, you can fine-tune them, you can enhance them, and you can guardrail them as you desire.

These models don't just operate on their own. They operate within what's called a harness, and that harness sits within what's essentially called a sandbox. These sandboxes are secure. They keep privacy and security and access control, and all of those capabilities are put into the sandbox.

These models are capable, excellent models that advance the whole industry. We want AI to diffuse into the market as quickly as possible, and open models allow for that.

Can intervention by the government ever be a good thing?

4. When AI regulation makes sense

Of course. Regulation is a good thing in many different industries, and you want to regulate applications. You have to be mindful about regulating technology. The raw technology itself is dual-use, and so we want to make sure that you regulate how it's used.

Maybe you want to regulate the service itself. Maybe you'll regulate how it's used in medicine and how it's used in transportation. You should regulate how AI is used in autonomous vehicles. All these different applications of AI should be regulated. However, the technology itself ought to be able to advance as quickly as possible.

5. The rules for competing with China

Now you've said China is an adversary in AI. What should the rules of engagement be for the U.S. with an AI adversary?

The first thing is to recognize that AI has dual use, and we want to advance it as quickly as possible because when the technology is advanced, it's more safe and more secure because it works better.

However, we want to make sure that the researchers in all of these countries are able to have dialogue and cooperate whenever possible so that we can advance the technology safely. I think an open dialogue is really important, and that should be foundational.

6. Why trust Nvidia on China and the market?

Your stock price depends partly on sales to China. Why should we trust what you say about China and the market?

We should start with the fact that our China sales are approximately zero today. I've told all of our investors not to expect any China sales.

If both the United States—and the United States has already licensed it—and the Chinese government and the Chinese market would like us to return, it would be a great, great honor to be able to support the customers and the market.

And good for business.

And great for business. But until then, I would just assume that sales are zero.

7. Why Chinese AI models help Nvidia

That's Kimi, by the way. Now, Kimi—the very powerful Chinese model that really rattled the stock market—Microsoft is using and testing Kimi. The Information says Chinese models may even replace some of Microsoft's use of OpenAI and Anthropic. This Chinese competition is coming fast and furious. What should U.S. AI companies do?

These Chinese models are excellent. Open-source models that are excellent should be used. The market misunderstood the impact of DeepSeek the first time. It has misunderstood another Chinese model. It's misunderstood the impact of Kimi again this time.

First of all, with great AI open models, it's great for the whole industry. Obviously, if there's great AI, even if it's open, wherever it comes from, there will be more use. Whenever there's more use, you'll have to sell a lot more NVIDIA computers. We'll have to build more data centers, we'll have more services, and the technology will diffuse into more industries.

The starting point is that great models lead to great use, which leads to great growth. That's what happened with DeepSeek, and it's going to happen with Kimi.

There's also a misunderstanding that open models could be adversarial to the closed models. That's also got it wrong. The reason for that is because the most likely person to upgrade to a great model like Anthropic and OpenAI is someone who already uses AI and would like to have it more conveniently.

Maybe they'd like to have more services come along with Anthropic and OpenAI. Maybe the models are just so much better, which I expect them to be. I think the thing that we want to do is make sure that we have AI technology diffuse into the industries as quickly as possible and have as many people use it as quickly as possible.

Open models, extremely cheap models, and easy on-ramps to try AI—that's the best thing we could do.

Now, NVIDIA has its own open-source model, Nemotron. When we were making it, you were telling me it's at the frontier, not the frontier. Will it become the dominant U.S. open-source model?

8. Nvidia’s open-source AI strategy

We built it so that companies and industries that, for sovereignty reasons, regulatory reasons, privacy reasons, or very specific domain intellectual-property reasons, would like to build their own AI can do so.

I advise all of my customers to use OpenAI and Anthropic, and to use open and closed models and services as much as they can because it's easy and it works incredibly well. I use it. I use Perplexity, Anthropic, and OpenAI. Use these services as much as you can.

You only build your own AI if you must. It turns out that a lot of companies—whether it's financial-services or cybersecurity companies, or companies like ourselves—must build our own AI. Nemotron is built for companies who need to do that.

We don't have to be the frontier. We have to be at the frontier.

And how quickly could you push out the big boys, or compete with the big boys, at the frontier?

That's not our goal. Our goal is simply that if you need to build your own custom models and custom AI, we could be of service to you to help you do that.

With the right harness—these harnesses are, for example, Open Claw, Hermes, or Claude Code—it's a harness around the Opus AI models. The large language model is the brain. The harness, if you will, turns it into an agent and into a thinking, working agent that can help you do things.

With the right harness, NVIDIA's Nemotron is extraordinarily good. You could adapt and customize it for the certain skills that you have in mind, and it can be completely world-class.

9. Can open-source AI threaten OpenAI and Anthropic?

What are the economics for the proprietary labs if open-source models, not just Chinese ones, can produce tokens—the intelligence—more cheaply?

Fantastic. Really, really, what the world needs is more people who need AI. The only way that you need AI is because you tried it and realized you want better ones.

These free services and build-it-yourself AIs are not for everybody. However, once you've tried it and realized it's so clumsy, it's so much work...

So you’re saying that if I’m OpenAI or Anthropic, I should not be afraid of open models?

Not at all. I think what we need—what they need more than anything—is for the whole world to realize they need AI. The way that people want to have better AI is that they try AI today and realize it would be nice for somebody else to help me service it.

It’s no different than cloud services. It’s no different than search. It’s no different than databases in the cloud. We could run our own search, but what a hassle it is. We can, in fact, build our own computers. Why don’t we just use them in the cloud?

There are a lot of things we could do that are more cost-effective, but it’s, in fact, more convenient to use them when they’re at Anthropic and OpenAI. I would also argue that for a lot of people and a lot of companies, even if the model is free, it’ll still cost you more to operate it yourself. You might as well have somebody else do it for you.

So I think it’s just more accessible, and it’s going to be better. However you think about these closed models, I think they’re going to be thriving for years to come.

What’s your vision for the future, with open-source and proprietary models competing with each other?

10. Using open and closed AI models

The world is going to have closed and open models. We should use closed models as often as we can, and we should customize and use open models to build our own customized AI, if we must.

Of course, the world is large, and for many countries and companies, they must build their own intelligence. They can’t outsource their fundamental intelligence to a third party. I think you should apply, adopt, and rent intelligence from cloud services wherever you can. However, you can’t outsource your own intelligence. You should build your own custom AI.

The Palantir CEO went viral for his interview, warning about giving up your alpha—your intellectual property—to the frontier labs. You’re in a unique position: You both use the frontier labs and supply them. You use them for your own chip design, and they’re your most important customers. Do you worry about giving away your alpha, your IP, to an AI that may be using it to do its own optimized chip design?

Nobody should outsource their alpha. Nobody should outsource their intelligence. No country should. We should always import and rent services wherever we can.

11. How Nvidia protects its technology

How do you protect it, and where are you most vulnerable?

Well, in the areas that are domain-specific, proprietary, maybe sovereign, maybe secret, maybe regulated, I need to do it myself. A company or a country needs to do it themselves.

However, there are a whole lot of things that we do that are not related to that. Maybe it’s related to automating marketing, or maybe it’s really automating our legal department. In all of those many different use cases around the company, it’s really about enhancing the productivity of the core company itself. It’s not a specific domain or intellectual property of the company. In those cases, I would outsource as much as we can.

12. The most astonishing thing Huang has seen AI do

What’s the scariest or most awe-inspiring thing that you personally have seen AI do—where you said, “Shit, I can’t believe this is happening”?

I expect it to do almost every task that we can do. It wasn’t a far reach to imagine that AI could generate videos that are photorealistic and so lifelike. It isn’t beyond imagination that AI can generate text, and from that it can reason. From that, it can essentially write code.

The things that AI is doing are fairly consistent with what I expect it to be able to do. I expect it to drive cars, move around and manipulate things, pick up things, and essentially manufacture these computers autonomously. I’ve largely imagined all of those capabilities.

I think the part that the market, the industry, and society need to realize is that while we’re automating all of these different tasks, it’s actually increasing the number of jobs that the world needs. It’s very sensible to me. While people have said that AI is going to eliminate jobs and make it so that we don’t have to work at all, it’s exactly the opposite.

I’ll first give you the evidence: The number of radiologists in the world is increasing substantially—some 20%, I understand. The reason for that is because the study of radiology scans has become completely automated by AI, but as a result, they can see a lot more patients.

Because there are so many people who need care, the number of patients they can now bring to the hospitals has increased dramatically, which results in needing more radiologists. This is the same with paralegals. Apparently, the number of paralegals has increased by some 10%, and the reason for that is that while they’re using AI, the number of cases they can see has increased.

There are just so many cases backlogged that now they need a lot more radiologists, a lot more paralegals, to be able to see all these different clients. This is happening all over. The number of manufacturing jobs has increased some 50% in the last several years because these computers have to go into these AI data centers, which generate the tokens, the intelligence. Because that is in such great demand, the number of manufacturing jobs has gone up.

When you think about AI across the whole industry, jobs are being created everywhere.

13. What about AI gives Haung pause

One more on what blew your hair back. We know that Anthropic’s Mythos can break into a bank or an intelligence agency. What has surprised you, or what’s coming that gives you pause?

Well, it surprised me that people were surprised. An agent can generate code because it can understand software. If it can debug software, it can find vulnerabilities in software. Cybersecurity and finding bugs are not inconsistent ideas.

The fact that AI can generate and debug software suggests it could also be used for cybersecurity. The flip side is equally true. We need to make sure that open models are in the hands of cybersecurity experts around the world.

Open models enhance cybersecurity. Open models enhance safety. The reason for that is because we want massively distributed, self-diverse defense. We want to put open models in the hands of cybersecurity experts all over the world so that they can defend themselves and look for vulnerabilities. They can share ideas among communities.

If everything just becomes one single model, one single point of attack, one single source of failure, I think the world is much, much more vulnerable. The thinking that somehow we have to close models so that the world could be safer is exactly the opposite. We have to open the models so that the world can have self-cybersecurity defense.

At age 30, 33 years ago, you started NVIDIA, which was then a company bringing 3D graphics to video gaming. You’ve been CEO since day one. You now have the world’s most valuable company. It turned out that the same technology that could be used in video game graphics cards is used in AI chips.

14. Could Nvidia’s success happen again?

NVIDIA was worth $1 trillion three years ago. Now NVIDIA’s worth 5 times that—$5 trillion. Could a success story like that ever be done again, or are you a unicorn’s unicorn?

Anything’s possible in the United States. There’s a reason why it’s called the American Dream. If you’re willing to work hard, if you want to take chances and go do something unique, the environment, the conditions, the support, the people, and the brilliance that’s all around us make it possible.

I think this is the land of dreams. Almost anything’s possible.

Yeah. It may make it more possible. I always tell my nieces and nephews, “Great fortunes are going to be created on a laptop.”

All completely true. The number of startups being created today is growing extraordinarily fast. In the last 6 months, $300 billion has been invested into venture capital, into startups, just here in the United States—$300 billion. All of that is creating new jobs and new companies. It’s incredible.

Now, NVIDIA’s facing the toughest headwinds since the Great Recession, close to 20 years ago. You had a market rout, China rising, and the U.S. government looking to be more restrictive. Bringing your engineer’s mind to the job of CEO, how do you navigate all that?

15. Nvidia’s toughest headwinds

The world is changing every day, and you just have to stay alert. But I have every confidence that NVIDIA and the United States are going to continue to thrive.

The confidence comes not from hope; confidence comes from the fact that you’re surrounded by amazing people. The country fundamentally wants to succeed, and it creates the conditions for entrepreneurs and innovators to do great work. We’re surrounded by amazing brilliance.

We have an incredible local market. We have all the recipes, all the ingredients necessary for long-term success. I’m as bullish as ever, as confident as ever.

The United States was supposed to be 6 months ahead of China in AI. Not much margin for error. Now, the tech world has been rocked by Chinese models that rival OpenAI’s ChatGPT and Anthropic’s Claude. Has China caught up?

16. Has China caught up to the U.S. on AI?

Whether China has or not, I don’t think it matters. Let me tell you why. First of all, the idea that it’s a race with an endpoint—obviously, we’re going to continue to use AI forever. The United States is going to be here for a long time. China is going to be here for a long time.

We have to recognize that China has one of the largest populations of science and math students in the world. If they manufacture anything that's incredible, they manufacture math students incredibly. They manufacture more AI researchers than the rest of the world combined. The fact that China is going to be extraordinary at this is a foregone conclusion.

17. Why fearing AI could cost America jobs

The idea that we would hold China back is ill-conceived, and it's not going to happen. Nor are they going to hold the United States back. This is an incredibly innovative country. We have brilliant minds from all over the world. They come here because of the American Dream. Immigrants, American-born, you name it. We have great universities and great industry, and so I have every confidence that the United States will continue to thrive.

The most important thing that we have to do with AI is not to scare our industries, not to scare our society into not using AI. AI is not going to destroy all of our jobs. Someone who uses AI is going to take our jobs, and so we have to make sure that we adopt AI, diffuse AI into the industries as quickly as possible. Anything we do that limits that, anything that we do that prevents that from happening, whether it's because we scare people or talk all day about the dooms of AI, I think all of that is ill-placed and it's not helpful to the United States.

So long as our industries take advantage of AI, we're going to do incredibly well.

18. Huang rejects the AI doom narrative

On the doomers, did some of your fellow moguls over-index on warning people, talking about what could come?

I think it's fine to warn people. It's even better to warn people with a solution, and it's absolutely inappropriate to make things up. The fact that this is going to be the end of humanity is complete nonsense. The fact that this is going to destroy half of the American jobs is complete nonsense. All of the facts, all of the evidence, point exactly to the opposite.

Intuition and good wisdom would tell you that it's exactly wrong, that AI will increase productivity. When productivity is increased, it creates opportunities. Look, over the arc of time, society has become more and more productive as a result of technology. It has created more jobs, not fewer. If it created fewer jobs, we'd have 100,000 jobs in the United States, and that's not true.

It's created new industries, new opportunities, new markets, and new applications. All of that increases the amount of opportunity and growth. I think the early evidence, several years into it—it's now 15 years into the era of AI—is abundantly clear. We're creating a whole bunch of manufacturing jobs because AI requires industrial might.

It's created jobs in the energy sector, in the chip sector, in the infrastructure sector—land, power, and shell. It's created a whole new slew of AI companies, and of course the industries that take advantage of AI are also growing. AI is creating all kinds of jobs.

Last one on China. You took a very realistic point of view about the strength that China will have. So are OpenAI and Anthropic, 2 of your best customers, in trouble?

19. Are OpenAI and Anthropic in trouble?

No. Not even a little bit. Both companies are thriving. They're going to continue to grow. They're going to go public. These 2 companies will be the most successful IPOs in human history. How is it possible that 2 companies that are only a few years old are worth $1 trillion?

The idea that AI is being diffused abundantly—free AI being diffused all over industry, all over society—so that if someone would like to have an off-the-shelf, incredibly built service instead of making their own AI, instead of hosting their own AI, they have that choice, I think it's going to be fantastic. The more AI is used, the more people use AI, the more people rely on AI, the more opportunity there is for OpenAI and Anthropic. So there's no scenario where China runs U.S. companies off the road.

Axios

So there's no scenario where China runs U.S. companies off the road?

No, no, no. Zero possibility. The American tech industry is too vibrant, too smart, too innovative, and moves too fast. Whatever competition there is around the world, bring it on. The United States is ready.

20. What Wall Street gets wrong about AI buildout

Axios

After Kimi, the powerful Chinese model, dropped, investors dumped Nvidia shares. Chip stocks are down 18% in the past month. What's Wall Street getting wrong about the future of the AI build-out?

Free AI should be great for hardware. Free AI should be great for chips. Free AI should be great for data centers. I think the market just got it wrong. They got it wrong with DeepSeek. As you recall, when DeepSeek came out, our stock price was down some 30%. I think this is exactly the same thing.

What this industry needs is great AI. The fact that Kimi 3 is so extraordinary, Shen is so extraordinary, Qwen is so extraordinary, and Nvidia's Nemotron is so extraordinary—all of these models that are open are great for the industry, including OpenAI's GPT-5.6, which is incredible. Of course, Codex is incredible. Claude Code is incredible. All of these applications that are doing incredible work—that are useful AI—useful AI has finally now arrived.

Useful AI is also profitable AI. When something is useful, people desire to use it. All of this is good. Closed models are excellent. Open models are necessary, and both are great for the industry.

Axios

Some investors think that AI has peaked. What do you know about the technology and its forward direction that you can tell them?

21. Why the AI buildout has barely begun

It's impossible that AI has peaked because the diffusion of AI into society and into industry has barely begun. What the world is seeing is that this new industry requires CapEx, unlike the last IT industry, which was very CapEx-light. The software industry was very CapEx-light, which is the reason why software companies have such high gross margins.

Axios

Capital expenditures. You have to build things.

That's right. In order to make things—in order to produce modern software—you need a computer like this to generate the software. When AI is doing the coding, it's this machine that's doing all of the processing of the large language model. It's doing the reasoning and the thinking. It's using all kinds of tools.

To do that, tokens have to be produced. Math has to be produced in this machine. This machine generates the intelligence that we see on our screen when words are coming out or images are coming out—or, in the future, when proteins and chemicals are being produced and robotic systems are articulating a maneuver, maneuvering. All of that is because this machine is generating the mathematics behind it.

The new IT industry, the new software industry, and, in the future, every single industry will be CapEx heavier—not CapEx-heavy, but CapEx-heavier. Most industries today, most intelligence industries today, are CapEx-light. That's simply going to change. But what you get as a result is incredible intelligence, incredible productivity, and incredible growth.

The growth, profitability, and prosperity that's going to come as a result of intelligence being manufactured is going to more than make up for the CapEx that we're investing in today. We are laying the foundations, building the infrastructure—the largest industrial infrastructure build-out in human history—so that we can build a foundation of industries in the future for the next generation. This is just the very beginning.

Axios

Semiconductors, as you have lived, have always been a boom-and-bust industry. Are we due for a bust?

22. Is the semiconductor industry heading for a bust?

No, not for a while. The reason for this is that this time is different. This is not demand-driven. It's not cyclical. It's not seasonal. This is industrial-driven, meaning the fundamental technology of computers is changing. We need a whole new layer of infrastructure in the world.

We have energy. We have the internet. We have roads. We have railroads. We now need AI—the intelligence layer, this infrastructure layer built on top of all of that. This layer needs chips. I believe we probably need our semiconductor industry to be somewhere between 5 and 10 times larger than it is.

Axios

Over what period of time?

Over the course of the next 10 years. It's way too small today, which is the reason why memory is in such short supply. Storage is in such short supply, optical interconnects are in short supply, and packaging is in short supply. Of course, TSMC's foundry capacity is in short supply.

Everything is in such short supply because it's not cyclical in nature. It's not demand from consumers. It's not because of anything like that. It's fundamental infrastructure-related. The whole industry is short. I think we need to build up the whole industry several more times.

23. When Nvidia starts worrying about AI spending

Axios

Your customers are selling hundreds of billions of dollars in debt to pay for your products. When does the party stop? When does your finance team start to worry about these deals?

We're not worried about these deals because of several factors. One, these companies are extraordinary companies, and they generate a lot of cash. What they're all doing, what they all see simultaneously, is that there's a fundamental change in how computing is done. It's a compute-platform shift.

Whereas computers used to be all content that was prerecorded—we wrote it in advance, we recorded it in advance, we took a picture in advance, we wrote a story in advance—in the future, all of that is going to be foundational information that we generate intelligence on top of. Instead of search, you simply ask a query. You have a question, and it gives you answers. Instead of combing through all these different websites, the AI will generate the answer for you.

24. When will AI investments pay off?

This future is a whole new way of doing computing that's fundamentally different from the past, and we need a lot more computers. Now, the reason why, a couple of years ago, some of the concerns were, “When is the ROI coming?”

Axios

Yes.

We now know that AI is profitable. It's the reason why Anthropic is growing so profitably and so incredibly fast. AI is both useful. That's why we're using it for coding. AI is profitable. That's why they want to manufacture so much more.

OpenAI wants to manufacture so much more because these coding agents are incredibly profitable. They're doing useful work for very high-paying jobs. So many companies like ourselves are happy to pay hundreds of millions of dollars a year so that we can use these AI services to augment our coding capability. We become more productive, we innovate more, we create more goods, we create more productive growth, and we become more profitable, which pays for more AI.

So that flywheel has now started. I think the industry has now arrived at a very critical juncture and inflection point, if you will, where we now have profitable AI because we have useful AI. Because of that, we want to build a lot more AI.

25. Why AI tokens could become more profitable

Axios

Here's the crux. You've defined a token as the fundamental unit of artificial intelligence. What convinces you that a token is going to keep getting more profitable?

The token is what's called an embedding. It embeds knowledge; it embeds intelligence. That number—that token that's being generated by this machine—is not a static number. It's not like pi. That number encodes smarter and smarter intelligence over time.

That single number is not a single number; it's a whole series of numbers—trillions and trillions of numbers. Those numbers are getting smarter and smarter over time. When they get smarter over time, they become more useful, more valuable. When they become more valuable, people will pay more for them.

I know that for most people, the concept of numbers being valuable is hard. But this token, these numbers, embed intelligence, and the intelligence is getting smarter and smarter all the time.

Axios

Every industrial build-out in history has seen a bubble of overinvestment. What is this generation's bubble danger?

26. The real danger of an AI bubble

The bubble will come someday. It's just not today. This is in the very beginning of the build-out.

Axios

Someday—five years, 10 years?

It's very unlikely in the next five years.

Axios

5 to 10?

Yeah, it's very unlikely in the next five years. We'll see about the 5 to 10 years. It just depends on how fast we can build.

The challenge, of course, is that the rate at which we can build the infrastructure is limited by physical things. Today, the industry could build faster, it could run faster, but we don't have enough chips, we don't have enough memory, we don't have enough land or power, and we don't have enough construction workers to build the data centers. We're basically constrained in every single direction, in every single way.

That constraint is good. That constraint is what holds the system back. So that gives us plenty of time to build out this infrastructure. On the one hand, there's a great desire to invest the CapEx. On the other hand, the ability for that CapEx to land and turn into productive supercomputers like this, into AI factories, is delayed because of all these different constraints.

I think it's going to push out the time when supply exceeds demand for some time.

27. Would Jensen Huang use Chinese AI model Kimi?

Axios

Would you use Kimi, the Chinese open-source model?

Of course. It's smart. You still have to make sure that you fine-tune it, safeguard it, guardrail it, put it inside the appropriate guardrails and the appropriate sandboxes, put security around it, and give it access control.

Just like all software, we download software from open source all the time. One of the most important pieces of open-source software in history is the operating system. It controls the computer. The operating system literally controls humanity today. It's the control fabric around the whole world's digital systems.

It is open source, based on Linux. It has the opportunity for millions and millions of people to study it, interrogate it, and test it to make sure that it's hardened. As a result, we can trust it. Open source is a fantastic way for us to harden and trust the infrastructure below us.

We really need to have open-source models. It's essential for society, it's essential for industry, and it's also the safest and most secure way to go forward. I think it's fantastic that Kimi is great, and that there are American open-source models and European open-source models. There are also Chinese open-source models and Japanese open-source models. I was just in Japan. There are Indian open-source models. I'm working with many of them there.

There are going to be open-source models from all over the place. That openness is fantastic and great for safety, great for security, and great for national security.

28. Why Asia is more optimistic about AI

Axios

You're arguably the most beloved AI CEO. People line up for your autograph when you're in Asia. Paparazzi follow your car. People want you to sign their clothing. Why is the view of AI so different in Asia than here?

I think maybe because the doomers spend too much time theorizing about these science-fiction outcomes. Maybe it makes them sound smart.

Axios

Are you talking about some AI CEOs?

29. Are AI CEOs scaring the public?

Well, I think leaders ought to be thoughtful about doing this. Listen, if you want to warn the world about the incredible capabilities of this technology, I think that's been achieved. We should dedicate our time to, of course, how we make the technology safe. That's our responsibility as technology leaders. Telling everybody that the AI technology has to be safe—you're just reminding yourself that you have to make it safe.

So we have to make sure that it's safe. We have to make sure that people understand the capabilities of this technology. We have to advance it in a way that's secure, safe, working, functional, and used in smart ways. It's our responsibility to do that.

I think, first of all, the warning is well-heeded. We have to make sure that regulators and policymakers are able to understand the nature of this technology. On the one hand, we want to encourage the American people to use this technology.

30. Why America could get left behind

We lead this industrial revolution not because we invented all the technology, but because we applied it with such enthusiasm that this industrial revolution led to the United States. We need to make sure that we apply this technology with great enthusiasm in health care, which we know. A doctor at John Hopkins just wrote that AI technology is helping advance the discovery of cancer a lot more quickly.

We know what it's done for radiology. We know what it's doing for software engineering. Software engineering is the fundamental engine of intelligence today and the fundamental engine of automation and productivity today. We know all of the things that AI is helping us do. That's incredibly good for us.

We know what it's able to do in helping us advance climate science and understand the climate better. Because of AI, we're able to upgrade our national power grid. Because of AI, we're able to invest in sustainable energy without government subsidies. All of this is because of this particular industry. I think we ought to be much more enthusiastic about it.

Let's help the United States realize that the only way we get left behind is if we don't apply the technology. Forget who contributes and invents the technology. The fact that we don't apply it—that's how we're going to get left behind.

Let's make sure that we remind ourselves of how we became the United States in this modern industrial revolution and that we apply exactly the same enthusiasm, the same spirit, the same American spirit, so that we can lead the next one.

31. Violence and hostility toward AI

Axios

Some of this burning opposition to AI has even turned violent. How are you starting to think about your own safety and security?

I'm in a safe place. I don't feel endangered. Mostly, I think you have to advance the technology safely, make it a contribution to society, always think about other people's interests, and make sure that we develop this technology in a way that helps society and industry broadly.

32. Why AI could close the technology divide

One of the things that I like to remind people is that, of all the technologies we've ever invented, this one closes the technology divide more than any other, and you can prove it. This is the first amazing technology in history where anybody can use it. That is the reason why so many people have used ChatGPT.

If you don't know how to use AI, go up to an AI and say, “I don't know how to use AI. Teach me how to use AI.” AI will teach you.

Axios

Improve my prompt.

Yeah, whatever it is—improve my prompt. Exactly. Whatever you imagine, whatever you dreamed, whatever you want to be better at, just tell AI and give it a chance to take you along a journey.

Goodness, I use AI every day to advance my knowledge, help me solve problems, and help me understand things that are outside of my domain of expertise. AI has been a great coach, a great adviser, and a great tutor to me. I think it should be for everybody.

Axios

And yet there is this hostile national mood. How worried are you that the next Democratic presidential candidate will be an anti-AI socialist?

33. Could anti-AI politics shape 2028?

We need to help people understand what AI is and what AI can do. The first thing I would remind everybody is that AI is creating an enormous number of jobs, not taking them away.

When we tell stories about AI destroying jobs, of course people are going to be anxious. Nobody says, “You know what I want more than anything? I want to invest in something that would destroy jobs.” Nobody wants that.

34. Huang says the jobs narrative is wrong

I think the narrative is wrong. The rhetoric is wrong and hurtful. I think the people who are talking about this have, in fact, got it fundamentally wrong.

Axios

So you’re saying the industry needs to tell a different, better, consistent story?

Yeah, exactly. But it’s not just a story. It’s ground truth. Just tell people the facts. The facts are, we’re creating millions of jobs.

Axios

So the industry needs to talk more and differently about the reality?

That’s right. Don’t create a narrative. Don’t create a story that is made up. It is made up that there’s going to be a singularity. It’s made up that somehow we’re living in a simulation. These are all made-up stories.

I think it’s a fun story. I don’t mind listening to it, and I even enjoy the narrative when it’s told by so many of those leaders and my friends. But we just all have to be thoughtful and careful.

Axios

So you’re saying you’re scaring people, and wrongly?

Yeah.

Axios

Separately from jobs, you’re talking about the sort of creepy elements of the technology, and you’re saying there’s a way to inoculate against that?

That’s right, because they’re just made up. They’re science fiction. They’re Hollywood. It’s not true. It’s not true. The closest thing to true AI is R2-D2 and C-3PO. Who doesn’t want R2-D2 and C-3PO?

Axios

You’ve developed a warm relationship with President Trump. You text, and he’s very responsive. What’s it like to communicate with the president like that?

35. Huang’s relationship with President Trump

He’s smart. He remembers everything. Goodness gracious, he sure knows numbers. And he’s the only president that has ever remembered H100, H200, NVIDIA’s Blackwell. He knows our next generation, Rubin. He has just an amazing memory.

We’re sitting here in a new factory. The first time I met him, he said he wants to restore the manufacturing capacity of the United States. He wants the United States to be reindustrialized. He wants a secure and resilient supply chain. He wants to move semiconductor manufacturing back to the United States.

Literally, we’re sitting in a plant that is a direct result of the first conversation I had with President Trump. This was in Fort Worth, Texas. We’re in Fort Worth, Texas.

He is single-minded about creating jobs. He is single-minded about reshaping the American economy and the labor workforce. He was in every single conversation I ever had with him. He is single-minded about that. He is single-minded about making sure that the United States economy is thriving, because he knows that if America’s rich, America will have the mightiest military, because you’ve got to fund it somehow.

And so having the United States have a vibrant ecosystem, a vibrant economy, and be a rich nation is foundational to the safest nation. He understands that very clearly, and he’s just so single-minded about it. It’s really inspiring.

36. What the Trump administration gets wrong on AI

What does the administration get wrong on AI?

I think that regulation of the technology needs to think about the unintended consequences of regulation. We need to think about AI not in the context of some kind of footrace in a 100-meter dash. The narrative that whoever gets to the 100-meter dash, whoever wins that, wins it forever, is nonsense.

And here you’re talking about the U.S. and China?

That’s right. It’s complete nonsense. The fact that all of those stories are made up is scary.

In Washington, D.C., where do you think it’s scaring the White House?

It’s scaring policymakers. And there’s a reason for that, because some of the companies hope that the government would be helpful in creating regulations to their advantage.

I think that we ought to compete. I think we ought to openly compete. We ought to recognize that this is very important technology, and the ultimate win is when society uses it. The ultimate win is when industry uses it. It has nothing to do with who invents it.

We didn’t invent electricity. We didn’t invent most of the technology. We didn’t invent manufacturing. We didn’t invent most of the technology that’s led to today’s Industrial Revolution. And yet the United States applied it faster, more enthusiastically, and with more ingenuity than any other country. It led to the United States of today.

I think we ought to let the American spirit thrive. We ought to let American industry run. Of course, we have to be mindful about regulation and make sure that the technology is safe. But to think that there’s going to be one winner, and whoever crosses a finish line 17 days from now—that’s complete nonsense. That’s completely made up.

Fun question: Who’s someone in the president’s Cabinet who’s smart and good to work with on building the future?

37. Who Huang trusts in Trump’s Cabinet

I love working with every one of them. Susie’s incredible—Susie Wiles, the chief of staff. Secretary Bessent, Secretary Lutnick: these are incredible people. They dedicate themselves to President Trump, and they want to see his presidency succeed. And we all do.

He’s my president, and I want my president to be successful so that our country can be successful.

There definitely are signs that they’re getting more restrictive. We’re looking at heading that way. Are you worried about that?

38. His warning to the White House

Sure. I would just add a caution, and this is to the administration, to the president, and to any policymaker—all policymakers: before you listen to all the rhetoric, before you listen to all the stories, before you manifest some kind of artificial intelligence future that is just not true, that is not grounded in fact. It’s completely science fiction.

The technology is very important. Of course, the technology is powerful. I’m one of the people in the world who builds the technology. The idea that somehow AI has consciousness—that’s all made up. It’s all complete nonsense.

And it sounds like you’re basically afraid the administration is going to fall for it and overcorrect?

Overcorrect. Overcorrect. Because if they fall for these narratives, I think they ought to be sure to talk to more people. Talk to many CEOs. Talk to many scientists. Don’t talk to 1 or 2. Don’t make a knee-jerk reaction.

Tomorrow is not some impending, imminent day that you have to make a decision. Be more informed. Take your time. Of course, be deliberate. Of course, be focused. Of course, be purposeful. But be informed, and give yourself lots of opportunities to talk to people.

39. Should the government own part of Nvidia?

President Trump has floated the idea of the government taking stakes in leading AI companies. You run one of the most valuable companies on Earth. If President Trump called you and asked you for an equity stake, what would you say?

It’s unnecessary. The United States has an equity stake in NVIDIA. It has a fantastic stake in our success. We paid $10 billion worth of taxes last year, and we’re going to pay a lot more taxes this year. We create lots of jobs, and we generate lots of taxes.

I think that the more successful American companies are, the richer these companies are, the more taxes they generate, the more jobs they create, and all Americans benefit. And don’t forget: most Americans today are in the stock market. When the stock market is up, everybody benefits.

The United States already has a stake in all of these great American companies. We just want to make sure that they continue to thrive and be successful.

So you would rule it out?

I wouldn’t advise it now. It’s unnecessary.

Do you think we’re ready for Anthropic’s powerful Mythos to be available to everyone? Are systems sufficiently hardened?

40. Should powerful AI be available to everyone?

It should absolutely be available to everyone. It’s their responsibility to make sure that it’s hardened.

So you’re saying all users, not just selected institutions as it is now?

That’s correct. They have to make it available for people, and it’s their job to make the technology hardened, safe, and secure. Whatever jailbreaks or vulnerabilities there are in the software, as soon as they find them, they ought to fix them. That’s the nature of software.

And look what happened. It was released somehow. It was a jailbreak, and it was able to do things that maybe they were trying to guardrail it from doing. But everything was fine. You and I are here having a conversation.

Not that I’m encouraging it, but that’s the job of companies that are supposed to identify vulnerabilities and patch them up as quickly as possible.

So this is a message to both the White House and Anthropic that you think Claude Opus should be available to all?

Mythos should be available as a service. And remember, just because Mythos is not available, open models are available anyhow.

I think, let Anthropic run. Let them continue to advance the technology that they created. Let it be put in the hands of as many companies as possible so we can all use it and benefit from it. Holding Anthropic back is not in the benefit of the United States. Let them run.

41. Can AI companies learn from rival models?

Should open-source model companies be allowed to distill closed models in violation of terms-of-service agreements, then resell their highly capable open-source models in the competitive marketplace?

It depends on the terms and conditions of the service. If the service provider is unhappy about some company taking advantage of or exploiting the terms of their agreement and their license, they ought to reach out to that company. There are many ways to deal with that today. We have a lot of conventional ways, using laws and regulations, to do so.

However, distillation—learning from AI, learning from other sources of knowledge—is fundamental to intelligence. We’re constantly learning from other people. I’m learning from you, just from the questions you’re asking, and you’re learning from me. All day long, we’re learning from each other.

And so AI has to learn from something. The original AI, no matter whose it is, whether it’s a closed model or not, obviously scraped the internet of all of the previously learned knowledge. And now AI is generating more content than humans.

In another few years, the internet would be 99% AI-generated content, and that content is generated by some AI. And so you're constantly distilling the intelligence of some other AIs anyway. The idea that AI can learn—that's a good thing. We want AIs to be smart. We want every AI to be smart, and a smart AI is a safer AI.

I think the important thing is to split these 2 ideas. Should AIs be able to learn as much as they possibly can from as many sources of knowledge as they can? The answer is yes. Can AI violate privacy? No.

Can AI violate the terms of an agreement? The answer is no. If somebody violates a term of an agreement, contact that company. Take care of it.

42. AI investment in Israel and the Middle East

You have a lot of employees in Israel. How do you think about the Middle East AI build-out in the context of the war?

Well, first of all, I'm concerned about my families in Israel. I have 6,500 families there, and they've been in the middle of conflict for some time now. I'm happy that they're well, I wish them to continue to be well, and I look forward to when the conflict ends.

The UAE is investing in the Middle East and AI. I think that's terrific.

Are you still bullish on the UAE as a giant AI player?

43. Why Huang remains bullish on the UAE

Yeah, they have a vision to be an AI hub someday. Building upon their natural resources of oil and energy, they have the opportunity to reinvent themselves into an AI hub. I'm proud of them. I'm delighted that they're interested in this, and I'm delighted by their vision.

The UAE is already so modernized. It's a place where smart people go, and they're thriving. They're building new companies and new innovations. This is going to be their next source of industrial strength, and I'm looking forward to that. It's great for them.

I'm about to fall through the floor. I want to be respectful of your time. Might I ask you 1 question about jobs and then a couple of questions about life lessons? Is that okay with you?

I've got all the time in the world.

44. Why Nvidia wants to stay relatively small

Very kind of you. I'm going to ask you about that because I know that is one of your theories. You run one of the most valuable companies in the world, and you have 50,000 employees. You've said that in 10 years you may have 75,000 employees—quote, “as small as possible.” Many of the other tech giants are in 6 figures. How do you do that?

A strategy. Strategy involves utilizing your resources as efficiently as possible. Your goal is to realize a future vision with the limited resources that you have and to use those resources wisely. Part of my job is to be strategic, to be mindful about the resources that I have, to apply them as precisely as possible to realize that vision, and to maximize the return on invested capital, if you will, speaking in financial terms.

But it's really about getting the most impact for whatever resources you have. That is the fundamental job of a CEO.

I have the benefit of starting this company when I was 30 years old, as you mentioned earlier, and I've been doing this job probably longer than any CEO in tech history. So I have the benefit of practice. Over time, this is my craft, if you will. This is my kung fu. That's what you have to do.

45. Which jobs and tasks will AI eliminate?

In every industrial revolution, some tasks and some jobs become automated, even if more net jobs are created. What are the jobs and tasks that will go away, like in any other industrial revolution?

If your job is the task, then it's very likely that when that task is automated, your job will be eliminated or changed. For example, customer service call centers—that is a task that is very likely to be highly automated. However, the management, the coordination, the improvement, and the evaluation of all of the AIs that are now handling the actual calls will likely continue to be a human-oriented task.

46. Don’t mistake your task for your job

Just remember that a job has a purpose, which includes many tasks. One of my favorite examples is the original example about radiology. Radiologists have the task of studying scans, but the purpose is to end human suffering, to help people understand their disease, to collaborate with other doctors, and to help identify the disease. The purpose doesn't change when the task has been automated.

Something that I tell early-career people in my life is, don't mistake your task for the job and the opportunity.

That's exactly right. Hey, look, today most of the people in IT, in technology, sit in front of a keyboard typing. If we're to take a picture of this and look at it 10 years or 20 years from now, it would look like the pictures we see from the old days of offices, with people sitting in front of IBM Selectrics typing.

And so we don't type. Typing is not the job. Solving problems. Innovating. Discovering impact. Creating value. Helping end human suffering. Bringing joy and delight. Whatever it is that your job purpose is doesn't fundamentally change because you don't have to type anymore.

47. The ChatGPT moment for robots

We don't have to type anymore. That's great. When do you expect the ChatGPT moment for robots, when they'll be used and useful in the everyday lives of ordinary people?

The ChatGPT moment in 2022 was not the moment that AI was useful. We had to wait another 4 years. The ChatGPT moment was when we saw that it was interesting, that it did something surprising.

If that's the case, then the ChatGPT moment of robots has already arrived. You could tell a robot, “Put the apple in the drawer,” and the robot will actually reason about the sequence of tasks that it has to do, including opening the drawer before it puts the apple in there. It will reason about that.

When somebody sees that for the first time—sees a mechanical robot actually doing that—I think it opens their mind about what the future of robotics can do. I think that the ChatGPT moment of robots has already arrived. Now the question is, when are we going to make it useful? If we make it useful in the next 3 to 4 years, I would not be surprised.

We've moved into the agentic era of AI. Won't that be the next era of AI?

48. A future with one trillion AI agents

Well, the agentic moment of AI—the capability—is now here. Now it's about the diffusion of the capability.

Very soon, the reason why we need so many more computers is because today we have a billion people using computers. We use these computers—like right now, when you're not talking, we're largely not using our computer. The computer is unused. So we have a billion people using computers. Maybe at any given point in time, only 100 million people are using their computers at the same time around the Earth.

Now, in the future, we will be supported by, augmented by, and surrounded by a whole bunch of agents who are helping us do things. Those agents are going to be using computers all the time. Agents are not going to become computers. Agents are going to use computers.

And so we're going to have 100 billion, a trillion agents that are running all the time—smart agents, lesser-smart agents, specialized agents, super agents, all kinds of agents running all the time. And so the number of computers we need is going to grow tremendously.

We'll finish with a rapid round on life lessons. And thank you very much for your generosity and for taking the time.

I enjoyed it, Mike. It's always great to see you.

I'm grateful. Thank you.

Likewise.

49. Why pain and suffering create greatness

You're a great American story. Born in Taiwan, sent to the U.S. at age 9. You went to a pretty rough-and-tumble boarding school, it sounds like. You've talked about how you were bullied, and you started NVIDIA at age 30. You said the key to greatness is plenty of pain and suffering. What is the pain and suffering in your career that resulted in the most benefit, or that has allowed you to create and lead this company?

Nothing. No great athlete just happens to be a great athlete. It's a lot of practice when nobody's watching. A lot of setbacks, a lot of losing. Just a lot of pain and suffering. That is really what elevates craft. It elevates character. It gives you confidence and resilience, allows you to handle very difficult challenges.

It's no different from when athletes say that somehow time slows down when you're in the middle of doing something, when you're in the zone, or when the pressure is high. I feel the same way. When the stress is at its highest, when the tension is at its highest, I kind of feel like time slows down. And the reason for that comes from practice.

It could just come from doing something over and over again, from managing your own stress, managing your own focus, and having the confidence to realize you can work it through, and from being surrounded by amazing people. All of that is a combination of that condition that leads to you being able to do something at a very high level. And so pain and suffering is necessary.

What would you say to 9-year-old Jensen, who is abroad, about coming to America? What should he be excited about, and what should he watch out for?

50. Huang’s message to young immigrants

America is the greatest country in the world, full stop. And it's the greatest country in the world not because we don't have our challenges, not because we don't have our disagreements. It's the greatest country in the world because of all those great challenges, because of all those disagreements.

We somehow manage to work it through, and through open discourse, through a vibrant system that allows freedom—freedom to speak up, freedom to innovate, freedom to start something, freedom to collaborate and work with people. The environment naturally attracts amazing minds. It naturally creates a condition for innovators and entrepreneurs and great researchers and great minds.

It creates the condition for anybody with desire to be able to make a great living. You could be a construction worker, a plumber, an electrician, or a scientist.

You could create. You can be an artist. You could create a great living. And so I think that this is an extraordinary country. I advise every bright mind around the world to find your way here. We need you here.

This country was built by immigrants. This country is going to need amazing immigrants in the future. They’re going to come here and join the amazing minds that are already here. And so we have 400 million amazing people who welcome the brightest minds and the greatest people in the future to come here.

You don’t wear a watch. And at this moment, that’s probably a good thing. Why?

51. Why Jensen Huang does not wear a watch

Because now is the most important time. I refuse to let Outlook manage my life, and I refuse to let a watch manage my life. If I’m too late, somebody will let me know. In the meantime, I’m 100% here. And so now is the most important time.

52. His perfect Saturday with his family

One fun thing: you’ve said that besides work, you spend all your time with your family. Take us through a Saturday with your family when you’re not working.

Perfect. Saturday

I wake up early and I play with the dogs, and shortly after, Lori walks the dogs. I start working and I work until lunch, and we have a great lunch. The dogs are asleep.

We work through to dinner, and the kids come over. We make dinner together. We’re all together and enjoy dinner, have a cocktail, and play with the dogs. And that’s a perfect day. And so if I could have that day every weekend—well, as it turns out, every weekend is exactly like that.

Wake up at the same time, play with the dogs, hope to see the family, hope to see the kids come over. That’s it. That’s the perfect day.

Jensen Huang, thanks for a memorable, meaty conversation.

Thank you, Mike. Great to see you.