Extended interview: Nvidia CEO Jensen Huang on fears about AI

Summary

• Huang assigns “0% chance” to AI ending the world by 2030 and calls the doom narrative “irresponsible” and “not grounded in science.” His mechanism for the anxiety: labs spent 10–15 years in research and “in the last six months practically the industry has shifted to high-production ramp mode” — token generation now “incredibly profitable” — which forces engineering resources toward verification and safety, the same maturation aviation went through.
• His regulatory position is the tradeable headline: no new AI laws, and he alleges the labs publicly asking for regulation actually want relief from existing liability. “Read between the lines… they’re actually not asking for more laws. They’re asking to be relieved of the laws we do have.” He cites lab cybersecurity incidents (one lab had four) already addressed by unauthorized-entry and product-liability laws: “Do the practical first and then we can talk about the theoretical.”
• He “completely disagrees” with Dario Amodei’s call to stop selling powerful AI chips to China. Vera Rubin is “shipping in volume production,” “several years ahead of China,” and Nvidia’s best technology is provided to American companies first — but every layer of what he calls the “five-layer cake” (while allowing there could be six: energy, chips, models, data and applications) should then compete worldwide: “Our market is not his to cede.”
• On pace versus China: “We should go as fast as we can, but not faster than we should” — no unsafe shipping. If China cures cancer or leads in sustainable energy, “it’s great for us”; America is “packing heat as a nation,” with tech one of three national treasures alongside finance and the military.
• Pressed on Trump calling AI-danger fears a “hoax,” Huang wouldn’t adopt the label — “I don’t want to speak for him” — but says doomsayers “must be doing it for ulterior reasons.” He confirms he’ll attend next week’s state dinner with Xi Jinping, where his message will be to “share best practices… maybe set global standards” on AI products.
• On the bipartisan data-center backlash: “We’re sorry we didn’t come talk to you sooner” — but he insists water consumption “is a myth now.” Recirculating hot-water cooling uses “less water than a swimming pool evaporating all year long,” and data centers should contribute to local grids and lower power costs, with new schools and “lower property taxes” among the community benefits he highlights.
• The investment frame: data centers are “AI factories” — “energy comes in, intelligence comes out” — powering a “multi-trillion-dollar industry” and American reindustrialization. Of a $100 trillion world economy, “some 15 trillion dollars of it would benefit from more intelligence”; “It’s not manufacturing atoms. It’s manufacturing electrons.”
• He breaks with fellow billionaires fighting California’s Proposition 40, even at a potential ~$8B cost over five years. “The fact that I can afford to pay $8 billion in taxes over five years is a privilege… I’m not afraid of paying taxes. I’m just afraid of being poor.”

Deep dive

1. The doom narrative is “not grounded on science” — it’s a lab-to-product growing pain

• Kent opens with the whistleblower claim that humanity could end within ten years; Huang’s answer is categorical: “that’s completely false,” and later, “There is 0% chance that’s going to be the end of the world” in 2030. The drama is “irresponsible” — though he’s careful to add “their concern is not wrong,” only the framing.
• His diagnosis of why the fear talk is peaking now: the labs went through a transition “this first part of this year” where AI became genuinely useful and token generation “incredibly profitable” — name-checking “OpenAI’s latest Astra” (as spoken), Claude Code, Cursor, and Cognition. As products mature and roll out, engineering shifts from capability to “verification, testing, evaluation, benchmarking” — like air travel, where more people work on safety than on designing new aircraft.

2. Don’t write new laws — Huang says the labs want relief from the ones that exist

• The sharpest claim of the interview: asked why he opposes an FDA/FAA equivalent for AI when Amodei and Altman ask for regulation, Huang points to lab cyber incidents — “one lab had four incidents… another one just had a couple more” — and says cybersecurity, damage-liability, and product-liability laws already address them. “Go and read between the lines… they’re actually not asking for more laws. They’re asking to be relieved of the laws we do have. And I think that’s a problem.”
• On trusting four or five CEOs with the technology: they’re “all young CEOs, first-time CEOs” against his 33 years, and he’s empathetic — “I believe in the people” — but “I don’t approve of deflections and distractions about the end of the world.” He says their engineers can address the incidents with secure sandboxes, containment, closer monitoring, and public disclosure so the industry can learn, the way cybersecurity’s defender community already works.
• The operating rule he keeps returning to: “Do the practical first and then we can talk about the theoretical. But don’t talk about the theoretical without doing the practical.”

3. China: as fast as we can — and “our market is not his to cede”

• Should America slow down? “We should go as fast as we can irrespective of anybody else.” If China cures cancer or invents sustainable-energy technology, “it’s great for us” — and besides, “we’re packing heat as a nation,” with the tech industry one of three national treasures alongside the financial system and the military.
• Asked about David Sacks’ warning that slowing AI damages the economy and military, Huang reframes: “We should go as fast as we can, but not faster than we should” — unsafe products already invite existing law, so the companies developing these technologies are “perfectly incentivized to do the right thing.”
• On Amodei’s “do not sell powerful AI chips” to China: “I completely disagree.” Vera Rubin is “shipping in volume production,” America is “several years ahead of China,” and Nvidia’s best technology goes to American companies first — but AI is what Huang calls a “five-layer cake,” while allowing there could be six: energy, chips, models, data, and applications. Every layer, from FedEx and Walmart as AI companies to the chip sector, should “go and compete and not cede entire markets. Our market is not his to cede.”

4. Trump’s “hoax,” and the message for Xi

• Kent presses twice on whether Huang agrees with Trump that AI-danger fears are a hoax. He won’t take the word — “I don’t want to speak for him” — but restates his own view: end-of-the-world claims “must be doing it for ulterior reasons… maybe it’s political… maybe just attention grabbing.” He does endorse Trump’s apply-existing-regulations stance as “very practical, very sensible,” and says the existing laws and regulations can handle the incidents under discussion.
• At next week’s state dinner with Xi Jinping, his message: with AI moving from labs to products, countries developing AI should “share best practices… maybe set global standards about product expectations” — cooperate “in every possible way” except military and national-security areas, which are “for the governments to sort out.”

5. Data-center backlash: “We’re sorry we didn’t come talk to you sooner”

• His first response to the rare bipartisan protest movement is an apology, grounded in autobiography — he grew up in Oneida, Kentucky, a town of 600 with 300 kids in his school: “if out of nowhere a data center were to crop up right next to the one grocery store… I think I would be alarmed.” Builders should engage communities long before standing up a data center; “it just happened so quickly… we didn’t get a chance to be polite.”
• Kent’s pushback — the real concern is water and energy bills, not aesthetics. Huang’s rebuttal: evaporative cooling is gone, so water consumption “is a myth now” — recirculating hot-water systems use “less water than a swimming pool evaporating all year long” — and data centers “should contribute to the grid,” making it more resilient and reducing power costs, not raising them.
• The community package he highlights includes community centers, new schools, baseball fields, roads, “lower property taxes,” bigger setbacks, and better-designed buildings. He concedes the next five years will “no doubt” use much more fossil fuel, but argues the current AI investment phase creates an opportunity to invest in every form of power generation — solar, nuclear fusion or fission, and hydro.

6. AI factories as reindustrialization — and $8B in taxes as “a privilege”

• His explainer of what data centers actually are: a new industrial layer after electricity (“power anything”) and the internet (“find anything”) — an “AI factory” where “energy comes in, intelligence comes out.” Of the world’s $100 trillion economy, “some 15 trillion dollars of it would benefit from more intelligence,” making this “a multi-trillion-dollar industry” wrapping every country and company.
• The framing he wants to land: “It’s not manufacturing atoms. It’s manufacturing electrons” — potentially the first chance in 100 years, or at least 50 years, to reindustrialize America, with electricians, fitters, and construction workers “enjoying such prosperity now” building trillions of dollars of AI factories.
• He also argues Nvidia’s success depends on safe deployment and describes his stance on Proposition 40, which peers are spending millions to fight: “The fact that I can afford to pay $8 billion in taxes over five years is a privilege… I’m not afraid of paying taxes. I’m just afraid of being poor.” The Denny’s dishwasher-to-CEO arc closes it: “I am the ultimate American dream.”