NVIDIA CEO Jensen Huang said in a 70-minute interview with Axios that Wall Street has misread the impact of Chinese AI models such as DeepSeek and Kimi, arguing that stronger models lead to more usage, more applications, more data centers, and ultimately more demand for computing infrastructure. In his view, that means more demand for NVIDIA systems rather than less. He also said the semiconductor industry will need to expand 5x to 10x over the next 10 years and that he does not see a bubble forming in the next five years.
Huang used the interview to push back on several popular narratives at once. He dismissed AI extinction claims and sweeping job-loss forecasts, argued that open models and closed models can reinforce each other, said NVIDIA’s sales in China are now "approximately zero," and described a future in which AI agents run around the clock at massive scale.
Huang says Wall Street got DeepSeek and Kimi wrong
Axios host Mike Allen opened with one of the most sensitive issues in the conversation. He referred to a Financial Times report that China was considering tighter export controls on AI models and semiconductors. He also noted that after Kimi emerged, NVIDIA shares fell sharply and chip stocks dropped 18% in a month.
Huang’s answer was blunt. He said the market got it wrong when DeepSeek appeared, and is making the same mistake again. Better models, he said, produce more use. More use means more NVIDIA computers sold, more data centers built, more services delivered, and more industries adopting the technology. That cause-and-effect chain sat at the center of his argument throughout the interview.
Asked whether Chinese models such as Kimi should be avoided, Huang said the opposite: "Of course you should use it. That’s smart." He said users can download a model, fine-tune it, enhance it, and place guardrails around it. The model runs inside what he called a harness, and that harness runs inside a sandbox with privacy protection, security controls, and access controls. He compared open AI models to Linux. Linux is open source, reviewed and hardened by millions of people, and trusted because it is constantly examined. He suggested open AI can be approached in much the same way.
Huang also argued that open models and closed models are not natural enemies. The people most likely to upgrade to premium systems from Anthropic or OpenAI, he said, are those who already use AI. Free and open models lower the barrier to trying the technology. Once users become familiar with it, some of them move on to paid services with stronger performance. In that setup, a bigger open-model ecosystem can also widen the market for closed-model providers.
NVIDIA’s China sales are "approximately zero," Huang says
When Allen asked about NVIDIA’s revenue in China, Huang gave an answer he said many people would not expect: "Our sales in China today are approximately zero." He said he has already told investors not to expect revenue from China. If the Chinese government and market welcome NVIDIA back, he said, that would be "a great honor." Until then, investors should assume China revenue is zero.
The input text also notes that "approximately zero" is Huang’s wording. It adds that public financial filings show U.S. export controls have reduced NVIDIA’s high-end chip sales into China to extremely low levels, but that the exact figure should be checked against company filings.
He calls AI doomsday claims "nonsense"
The sharpest exchange in the interview came when the discussion turned to AI risk. Huang said warning people about technology risks is fine, and better if the warning comes with solutions, but "making up facts is absolutely not okay."
He then rejected a series of common AI fear narratives in direct terms. "To say AI is going to destroy humanity is complete nonsense. To say AI is going to eliminate half of all U.S. jobs is complete nonsense. All of the facts and evidence point in the opposite direction."
To make that case, Huang cited several examples. He said the number of radiologists has risen by about 20% because AI can automate scan analysis and allow doctors to see more patients. He said the number of paralegals has risen by about 10% for similar reasons. He also said manufacturing jobs have grown by about 50% in recent years because AI data centers have to be built and chips have to be produced.
His broader argument was that higher productivity creates opportunity. Over the course of history, he said, technology has made society more efficient and created more work rather than less. He also aimed criticism at what he called doomsayers, saying too much time has been spent theorizing about science-fiction outcomes. If the point was to alert the world to the extraordinary power of the technology, he said, that goal has already been achieved. The next task is to make it safe, and he framed that as a responsibility for technology leaders.
Allen also raised the contrast between attitudes toward AI in Asia and in the United States, noting that Huang is mobbed by fans in Asia asking for autographs. Huang’s explanation was that people in Asia tend to embrace AI as a tool and an opportunity rather than treating it primarily as a threat.
The input text adds an important caveat: the figures Huang cited on radiologists, paralegals, and manufacturing jobs were spoken in the interview and were not sourced during the discussion.
Why he thinks a bubble is unlikely in the next five years
Allen pressed Huang on a familiar question: if every industrial revolution eventually develops a bubble, where is the risk this time?
Huang said a bubble will come one day, but not now. In his view, the buildout is still in its earliest stage. He gave a rough timeframe: not likely in the next five years, with the five-to-10-year window less certain.
The reason, he said, is broad supply-side constraint. The industry could build faster, but it cannot get enough chips, memory, land, electricity, or construction labor. He described shortages in every direction and on every front. Those constraints, he argued, are actually helpful because they slow the system down and buy time for infrastructure to be built. Demand is strong, but the physical ability to convert that demand into productive supercomputers is delayed. That pushes out the point at which supply might overtake demand.
Huang tied that view to a larger call on the sector. He said he believes the semiconductor industry needs to expand 5x to 10x over the coming decade. In his telling, this cycle is different from previous semiconductor cycles because it is not seasonal, not consumer-led, and not driven by a standard demand swing. It is being pulled by industrial infrastructure. The world already relies on energy networks, the internet, roads, and railways. Now, he said, it is building an AI layer on top of all of that, and that layer requires chips.
Allen also asked about the risk that customers are borrowing heavily to buy NVIDIA products. Huang said he is not very worried because these are strong companies with large cash generation. He added that the AI monetization flywheel has already started. AI is useful now, he said, and once it becomes useful it can generate revenue. Coding agents were one example he highlighted. Companies are willing to spend hundreds of millions of dollars per year to strengthen coding capability, he said, because those systems can perform valuable work in highly paid roles.
Huang’s token economics argument: smarter AI means more valuable tokens
One of the more technical parts of the interview centered on Huang’s view of token value. He said a token is an embedding of knowledge and intelligence. Unlike a fixed numerical constant such as pi, he said, the intelligence encoded in those numbers becomes smarter over time.
His logic followed from there. Smarter intelligence is more useful. More useful intelligence is more valuable. More valuable intelligence can command a higher price. That is why he said tokens will become more profitable over time.
Huang contrasted the AI era with the earlier software era. Traditional software was relatively light on capital, which helped produce high gross margins. AI software, he said, will be more capital intensive because producing modern software requires machines like the supercomputer sitting in front of him, systems used to generate intelligence. He said every industry will become more capital intensive, but the tradeoff is extraordinary intelligence, productivity, and growth.
He described the current phase in sweeping terms: the world is laying foundations and building what he called the largest industrial infrastructure project in human history.
On the question of whether AI has already peaked, Huang said it cannot have peaked because social and industrial adoption has only just begun. He also referred to $300 billion flowing into venture capital and startups in the United States in just the past six months, saying those investments are creating companies and jobs.
Trump, regulation, and the shape of the AI race
Huang’s comments on Donald Trump were unusually specific. He said Trump is smart, remembers everything, and understands numbers. Huang said Trump could remember NVIDIA chip names such as H20, H200, and Blackwell, and even knew that the next generation is called Rubin.
He also recounted their first meeting. Trump, Huang said, talked about restoring U.S. manufacturing capacity, reindustrializing the country, building secure and resilient supply chains, and bringing semiconductor manufacturing back to the United States. Huang said the Fort Worth factory where the interview took place was a direct result of that conversation.
Asked what mistake the government should avoid, Huang said he worries about overregulation and overcorrection. He added that some companies want the government to write rules that favor them. His preference, he said, is open competition.
He rejected the idea that the AI race is a 100-meter sprint in which the first mover wins permanently. In his view, the real measure of victory is whether a society adopts the technology, not who invented it first. He pointed to electricity and manufacturing as examples. The United States did not invent either one, he said, but adopted them faster and with greater enthusiasm.
Allen then asked how Huang would respond if Trump called and asked for an ownership stake in NVIDIA. Huang said there would be no need. America already has a stake through jobs, taxes, and broad participation in the stock market. He said NVIDIA paid $10 billion in taxes last year and will pay more this year.
Open access, distillation, and Huang’s view on model disputes
On whether Anthropic’s Mythos model should be opened to everyone rather than limited to select institutions, Huang’s answer was clear: "Of course it should be available to everybody." He said it is Anthropic’s job to make the technology safe, just as software companies are expected to patch vulnerabilities when they are found. Referring to a jailbreak incident involving Mythos, he said, "everything’s fine, you and I are still here talking."
On the question of open-model companies distilling closed models and reselling the results, Huang struck a more balanced tone. He said it depends on the terms of service. If a provider believes its terms were violated, it should contact the company involved. He added that there are many traditional legal tools already available to address those disputes.
At the same time, he stressed that learning from other sources is a basic property of intelligence. Early AI systems scraped the internet’s existing knowledge, he said. Now AI-generated content already exceeds human-generated content, and in a few years 99% of the internet could be AI-generated. In that environment, he suggested, AI constantly distilling intelligence from other AI outputs is not a surprising development.
Robotics has reached its "ChatGPT moment," he says
Allen asked when robotics will have its ChatGPT moment. Huang’s answer was that it already has.
He drew a distinction between the moment a technology becomes obviously useful and the moment it becomes broadly surprising to the public. ChatGPT’s debut in 2022 was a surprise moment, he said, a point when people felt the system was interesting and astonishing. It still took another four years for it to become deeply useful. Robotics, in his view, is now at that earlier stage.
He gave a simple example. Tell a robot to put an apple in a drawer, and it can reason through the sequence: open the drawer first, then place the apple. The first time people see physical robots do that in the real world, he said, it changes how they imagine the future of robotics.
As for when robotics becomes widely useful, Huang gave a three-to-four-year timeframe and said he would not be surprised if that happened.
A world of 100 billion to trillions of agents
Huang’s view of the agent economy was even larger in scale. He said roughly 1 billion people use computers today, but those machines are idle most of the time. In the future, he said, every person will be assisted by many agents that use computers continuously.
He described a world with "100 billion, trillions of agents" running all the time: smart agents, less-smart agents, specialized agents, and super agents. If that happens, the number of computers needed will rise dramatically because those systems will keep consuming compute around the clock.
He extended that vision to business structure as well, saying every company will need AI employees.
On company size, leadership, and his own working style
The conversation eventually returned to Huang himself. Allen noted that Huang founded NVIDIA at age 30 and that the company is now worth $5 trillion while employing only 50,000 people. Huang said that 10 years from now it may still have only 75,000 employees, because he wants it to stay "as small as possible."
He defined the CEO job as strategy: using limited resources as efficiently as possible to realize a future vision. He said he has been doing that work since age 30 and may be the longest-serving CEO in technology history. "This is my craft," he said.
On the idea that pain and hardship are central to greatness, Huang said he does not think of it as one defining painful event. He described it instead as a sustained condition. Great athletes do not become great by accident, he said. They get there through training when nobody is watching, repeated failure, pain, and sharpening of craft. He said that process builds character, confidence, and resilience. Under pressure, time can seem to slow down, he added, and that feeling comes from repeated practice.
Asked what he would say to his 9-year-old self arriving in the United States, Huang responded with an emphatic defense of the country. He called the U.S. the greatest country in the world and said its challenges and disagreements are part of what makes it strong, because they exist inside a system that allows free speech, free innovation, free entrepreneurship, and free collaboration. He closed that thought by saying the country was built by immigrants and will continue to need remarkable immigrants.
Allen also asked why Huang does not wear a watch. Huang answered in character: because the present moment is the most important time. He said he refuses to let a schedule or a watch manage his life. If he is late, someone will tell him. Until then, he is fully present.
Conflict disclosures included in the input text
The input text includes a conflict-of-interest note stating that Huang owns about 3.5% of NVIDIA, or about 860 million shares, and that his personal wealth is about $181 billion based on Bloomberg data from June 2026. That note says the interview topics, including AI industry expansion, rising chip demand, and lighter regulation, are directly tied to his financial interests. It also says readers should judge for themselves his core argument that stronger Chinese AI models are positive for NVIDIA because they increase chip demand.
The same note says Huang’s comment that China sales are "approximately zero" should be checked against financial filings, and that the employment figures he cited in support of AI job creation were not sourced in the interview itself.