NVIDIA CEO Jensen Huang on China’s AI, Trump, and the trillion-agent era: Wall Street misunderstands DeepSeek; the AI doomsday narrative is nonsense; chip demand will expand 5 to 10 times

Curated & Compiled by: DeepChao TechFlow

Program: Axios Behind the Curtain

Guest: Jensen Huang, NVIDIA co-founder and CEO

Duration: 70 minutes

Recording location: NVIDIA factory in Fort Worth, Texas

Conflict of interest: Jensen Huang is CEO and co-founder of NVIDIA, holding approximately 3.5% of the company’s shares, with personal wealth of about $181 billion. This episode covers topics including AI industry expansion, chip demand, and Chinese government regulation, all directly related to his personal financial interests. His position clearly leans toward promoting AI industry expansion and reducing regulation.

Summary

The CEO of the world’s highest-valued company sits in his newly built Texas factory and takes a stance on nearly every hot controversy in the AI industry. Huang spoke plainly: Wall Street’s fear of China’s AI models is a misunderstanding, AI doomsday talk is “nonsense,” and the chip industry over the next decade needs to expand by 5 to 10 times. He also said a bubble is unlikely to appear within five years. He further said publicly that Anthropic’s Mythos model should be open to everyone, and that OpenAI and Anthropic will be the “most successful IPOs in human history.” When asked about Trump, he said the president has an excellent memory and can recall chip model names like H20, H200, Blackwell, and Rubin. The only thing that concerns him is the possibility that the government could be misled by “sci-fi narratives” into overregulation.

Key quotes

“The market misunderstood DeepSeek’s impact, and then misunderstood Kimi again. Great models lead to more use. More use means selling more Nvidia computers and building more data centers. The starting point is: good models bring good applications, and good applications bring growth.”

“Saying AI will destroy humanity is completely nonsense. Saying AI will eliminate half of America’s jobs is completely nonsense. All facts and evidence point the other way.”

“China’s AI researchers are more than the rest of the world combined. China will become exceptional in this field—that’s a given. Want to stop China? That’s a stupid idea, and it’s simply not doable.”

“The semiconductor industry needs to expand 5 to 10 times. Everything is scarce today: chip shortages, memory shortages, and shortages of land, electricity, and construction workers. This kind of scarcity is actually a good thing—it gives us time to build infrastructure.”

“OpenAI and Anthropic will become the most successful IPOs in human history. How could a company worth one trillion dollars be only a few years old? Because AI is useful. And because AI is useful, it can make money.”

I. The AI shockwave: Wall Street misread it again

In the opening, host Mike Allen raised the most sensitive topic: FT reported that China is considering tightening export controls on AI models and semiconductors. Kimi—a Chinese model—launched, and Nvidia’s stock price fell sharply. Over the month, chip stocks dropped 18%. What exactly did Wall Street get wrong?

Huang’s answer was direct. He said the market misread it once when DeepSeek arrived, and now it is doing so again. “A good model brings more usage. More usage means selling more Nvidia computers. We need to build more data centers, provide more services, and the technology will permeate more industries.” He repeatedly stressed this logic chain: good models bring good applications; good applications bring growth; and growth requires more compute power.

On whether Kimi and other Chinese models should be banned, his response was equally straightforward: “Of course they should be used—it’s smart.” He explained that after downloading a model, you can fine-tune it, enhance it, and add guardrails. The model runs within something called a “harness” (a framework), and the framework runs inside a “sandbox” (a sandbox). The sandbox is safe, with privacy protection, safety controls, and access controls. He compared it to an operating system: Linux is open source. Hundreds of thousands—or millions—of people around the world audit it, test it, and harden it. So we can trust it. Open-source AI models work the same way.

He also mentioned a point that’s easy to overlook: open models and closed models are not opposites. “The most likely people to upgrade to Anthropic or OpenAI—good models—are those who are already using AI. Free models lower the barrier to trying AI. Once users get started, they naturally want better services, and then they will pay for closed-model vendors. So the more open models there are, the more opportunities there are for closed models.”

When asked about Nvidia’s China revenue, Huang gave an answer many people did not expect: “Our revenue in China today is approximately zero.” He said he has already told investors not to expect any revenue from China. If the Chinese government and market welcome them back, that would be “a great honor.” Until then, treat revenue as zero.

II. Where the AI doomsayers get it wrong

This was the most explosive part of the entire conversation. The host asked: Are some of your peers in tech overhyping the risks of AI?

Huang’s response was blunt. “Warning others is fine. Warning people with solutions is even better. But fabricating facts is absolutely not allowed.” He then called out several popular AI doomsday narratives: “Saying AI will end humanity is completely nonsense. Saying AI will wipe out half of America’s jobs is completely nonsense. All facts and evidence point the other way.”

He offered several concrete examples to support his argument. The number of radiologists increased by about 20%: after AI automates scanning and analysis, doctors can see more patients, and because there are so many people who need care, they end up needing more radiologists. The number of legal assistants rose by about 10%—the logic is the same. Manufacturing jobs grew by about 50% over the past few years because AI data centers need to be built and chips need to be produced.

“Intuition and common sense will tell you that AI increases productivity. Higher productivity creates opportunities. Look at history: technology makes society more efficient and creates more jobs, not fewer. If it were otherwise, the U.S. would now have only 100,000 jobs.”

He also aimed at the AI leaders. “Doomsayers spend too much time theorizing sci-fi endings. Maybe that makes them seem smart.” When asked whether he was talking about CEOs of certain AI companies, he didn’t deny it. Instead he said: “If your goal is to get the world to be wary of this incredible capability of the technology, then that goal has already been achieved. We should use our time on how to make the technology safe—that is the responsibility of technology leaders.”

The host mentioned that Asia’s attitude toward AI differs sharply from the U.S., and Huang said fans surrounded him in Asia and asked for signatures. His explanation was thought-provoking: “Maybe it’s because doomsayers spend too much time fabricating sci-fi endings. In Asia, people embrace AI as a tool and an opportunity, not as a threat to fear.”

III. The chip industry needs to expand 5 to 10 times; a bubble is unlikely within five years

The host posed a sharp question: every industrial revolution goes through bubbles—where is the bubble risk in this era?

Huang answered cautiously but clearly. “A bubble will come someday, but not today. We are at the very beginning of building.” He gave a timeline: unlikely within five years; it depends between five and ten years. The reason is comprehensive constraints on the supply side. “The industry can build faster now, but we don’t have enough chips, not enough memory, not enough land, not enough power, and even not enough construction workers. We are constrained in every direction and every aspect.”

He said these constraints are a good thing. “These constraints hold the system back and give us plenty of time to build infrastructure.” Demand is strong, but the ability to turn demand into productive supercomputers has been delayed due to all these physical limitations. That pushes out the point when supply exceeds demand.

More importantly, he shared his view on the overall scale of the semiconductor industry. “I believe the semiconductor industry needs to expand 5 to 10 times.” What about the time horizon? “In the next decade.” That means today’s chip industry is still far too small—far from enough to support the AI infrastructure buildout.

He explained why this semiconductor cycle is different from the past. “This time is different because it’s not demand-driven, not seasonal, and not consumer-driven. It’s driven by industrial infrastructure.” Just like the world needs energy, the internet, highways, and railroads, we now need an intelligent infrastructure layer for AI, built on top of all existing infrastructure. And that layer needs chips.

On the risk that customers might take on debt to buy Nvidia products, he said he’s not too worried. “These companies are excellent. They generate a lot of cash.” He specifically pointed to the launch of an AI profit flywheel: AI is useful, so AI can make money. Coding agents are extremely profitable: they do useful work in high-paying roles, and many companies are willing to pay billions of dollars per year to enhance their coding capabilities. “This flywheel has already started.”

IV. Tokenomics: The smarter AI is, the more valuable it is

This was the most technically deep part of the whole conversation. The host asked: What makes you confident tokens will become more profitable over time?

Huang’s explanation was vivid. “Tokens are a kind of embedding. What they embed is knowledge and intelligence. This number is not static—unlike pi. The intelligence encoded in this number will become smarter over time.” “When intelligence becomes smarter, it becomes more useful. When it becomes more useful, it becomes more valuable. And when it becomes more valuable, people are willing to pay more.”

He compared the AI industry with the software industry in the past. In the past, software was “light capital,” so software companies had high gross margins. But in the AI era, software will become “heavier capital,” because producing modern software that generates intelligence requires machines—like the supercomputer in front of him. Every industry will become more capital-intensive, but the payoff is incredible intelligence, productivity, and growth.

“We are laying the foundation and building infrastructure—this is the largest-scale industrial infrastructure buildout in human history.”

On doubts about whether AI has already peaked, his response was: “It can’t have peaked because AI’s penetration into society and industries has only just begun.” He also mentioned that over the past six months alone, there has been $300 billion invested in risk capital and startups in the U.S., and all of this has created new jobs and new companies.

V. Trump, regulation, and government shareholding

Huang’s assessment of Trump was surprisingly specific. “He’s smart—he remembers everything. Oh my God, he really understands numbers.” He said Trump was the only president who could remember NVIDIA chip model names like H20, H200, and Blackwell, and he even knew that the next-generation product is called Rubin.

He described their first meeting. “He said he wants to restore America’s manufacturing capability, make America re-industrialize again. He wants a secure and resilient supply chain, and wants to move semiconductor manufacturing back to the U.S.” He said the Fort Worth factory where they sat for that interview was a direct result of that conversation.

When asked what mistakes the government should avoid, Huang rarely showed anxiety. “What I worry about is overregulation, overcorrection.” He said some companies want the government to help craft regulations that benefit themselves. “I think we should openly compete.” He dismissed the AI race as a “100-meter dash” framing as “nonsense”: “Whoever reaches the finish line first wins forever? That’s nonsense. Ultimate victory depends on whether society uses the technology—not on who invented it. We didn’t invent electricity. We didn’t invent manufacturing. But the U.S. used them faster and with more enthusiasm—so the U.S. became what it is today.”

The host pressed: If Trump called to ask for equity in Nvidia, what would you say? Huang’s answer was clever: “No need. The U.S. already has equity in Nvidia. Last year, we paid $100B in taxes. This year, we will pay more. We create a lot of jobs and taxes. And don’t forget: most Americans are in the stock market today. When the stock market rises, everyone benefits.”

On whether Anthropic’s Mythos model should be open to everyone (it is currently only open to certain institutions), Huang’s stance was very clear: “Of course it should be open to everyone.” He said it’s Anthropic’s responsibility to ensure the technology is safe—just like any software: when you find vulnerabilities, fix them as quickly as possible. He mentioned the Mythos jailbreaking incident, saying, “Everything is fine—you and I are still here chatting.”

On the issue of open-model companies distilling closed models and then reselling them, his view was more balanced: “It depends on the service terms. If the service provider is not satisfied, they should contact that company. We have many traditional legal tools to deal with it.” But he emphasized that learning from other sources is a fundamental attribute of intelligence in AI. “AI must learn from something. Even raw AI is grabbing all existing knowledge on the internet. Now AI-generated content is more than humans. In a few years, 99% of the content on the internet will be AI-generated. You’re already constantly distilling the intelligence of other AIs.”

VI. The chatbot ChatGPT moment of robots and the trillion-agent era

The host asked: When will the robot’s ChatGPT moment arrive?

Huang gave an unexpectedly answer: “The robot’s ChatGPT moment is already here.” He explained that the ChatGPT moment is not the moment when AI becomes useful. When ChatGPT came out in 2022, it only made people feel “interesting, surprised.” It took four more years for it to become truly useful. Robots are now in that stage of “interesting, surprised.” You can tell a robot, “Put the apple into the drawer,” and it will reason out the task sequence, including opening the drawer first and then placing the apple. When you see a mechanical robot do it for the first time, it opens up your imagination of the robot’s future.

As for when it becomes useful, he gave a timeframe of “three to four years,” and said, “I wouldn’t be surprised.”

Regarding the Agent era, he painted a shocking picture. Today, a billion people use computers, but most of the time the computers are idle. In the future, everyone will be assisted by massive numbers of agents—agents will use computers around the clock. “We will have 50k, 1 trillion agents running 24/7. Intelligent agents, less intelligent agents, specialized agents, super agents—every kind of agent will be running around the clock. So the number of computers we need will increase dramatically.”

VII. The philosophy of a CEO’s craft and pain

In the final part, the conversation returned to Huang’s personal perspective. The host mentioned that he founded NVIDIA at age 30; today the company is worth $5 trillion, but it has only 50k employees. He said ten years from now it might only have 75k people, “as small as possible.”

Huang defined the CEO’s job as “strategy.” “Strategy is to achieve the future vision as efficiently as possible with limited resources.” He said he started doing this job at 30, which is probably the longest tenure CEO in the history of tech. “This is my craft. This is my kung fu.”

Regarding the theory that “pain and grinding are the key,” he corrected it by saying it’s not tied to some specific painful event; it’s more like a continuous state. “No great athlete becomes great by accident. It’s a lot of practice when no one is watching, a lot of failure, a lot of pain and grinding.” He said this grinding improves the craft, shapes character, and gives people confidence and resilience. “Like athletes say: when the pressure is highest and the tension is highest, time seems to slow down. I feel the same. It comes from practice—doing the same thing again and again.”

When asked what he would say to himself if he were nine years old and coming to the U.S., his answer was a passionate tribute to America: “America is the greatest country in the world. Period. Even if we have challenges, and even if we have disagreements, it’s precisely those challenges and disagreements that make us great. Through open dialogue, through a vibrant system that allows freedom—freedom to speak freely, innovate freely, start businesses freely, and cooperate freely.” He concluded: “This country is built by immigrants, and the future will still need incredible immigrants.”

On his habit of not wearing a watch, his explanation was very “Jensen Huang.” “Because right now is the most important time. I refuse to let my calendar manage my life. I refuse to let a watch manage my life. If I’m late, someone will tell me. Until then, I’m here 100%.”

Notes

2. Conflict of interest: Jensen Huang holds approximately 3.5% of NVIDIA shares (about 860 million shares), with personal wealth of about $181 billion (Bloomberg June 2026 data). This episode involves topics such as AI industry expansion, chip demand growth, and reducing regulation, all directly related to his personal financial interests. His argument that “good models bring more chip demand” is essentially saying that China’s AI models are favorable to NVIDIA; readers should judge for themselves.

3. About “China sales = ‘approximately zero’”: Huang claims NVIDIA’s sales revenue in China is “approximately zero.” Based on public financial reports, U.S. export controls have indeed limited Nvidia’s sales of high-end chips in China to very low levels, but “zero” is Huang’s wording; the actual number needs to be confirmed from financial statements.

4. About the data that AI creates jobs: The figures Huang cited—radiologists up 20%, legal assistants up 10%, manufacturing up 50%—are Huang’s oral quotations and are not labeled with sources in the conversation. Readers may cross-verify on their own.