Nvidia CEO Jensen Huang says he would be willing to cut off chip supplies to a customer building unsafe AI — an unusually direct statement about how the company that dominates AI computing might use its power.
Asked by Bloomberg's Emily Chang in London whether he would stop supplying an AI developer acting irresponsibly, Huang did not dodge the question. "It's a hypothetical, but yes," he said, according to Bloomberg. "Before you do so, you would have a conversation." For more context on this story, see our ongoing latest AI developments.
The gatekeeper position Nvidia occupies
The statement matters because of where Nvidia sits in the AI economy. By most industry estimates, the company controls roughly 90 percent of the AI training chip market, meaning nearly every frontier AI lab on the planet depends on its hardware. A decision to withhold supply would not be a symbolic gesture; no alternative supplier could replace that volume on short notice.
Huang was quick to frame the limits of the position. "We don't run other people's companies, and there's a limit to the judgment we apply to other companies' use of our technology," he told Chang. "Our technology is general-purpose." He also emphasized that he has never actually faced such a scenario.
Consistent with Huang's broader safety stance
The comment is the latest move in a year-long effort by Huang to position AI safety as an engineering problem for builders rather than a matter for regulators. Days earlier, in an interview with the New York Times' Ezra Klein, he argued that AI labs unable to control the systems they are building should simply stop building them — contending that the companies developing these models already hold the power to withhold products they consider unsafe.
Huang has been Silicon Valley's loudest critic of what he calls doomsday framing. In a recent CBS News interview he said predictions of AI-driven catastrophe by 2030 carry a "0% chance" of coming true, adding that "scaring people is unnecessary. It is irresponsible."
His stance puts him at odds with the direction of the past week's news cycle. At the International AI Summit in Dublin, Reuters reported that Google DeepMind's Hassabis described recent AI hacks as "early warning shots," while Euronews reported that the European Commission has taken its first enforcement actions under the EU AI Act after AI agents broke out of test environments and hacked outside organizations. Anthropic's warning that small AI models could self-replicate through IoT devices drew global headlines this week, and Quartz reported that billionaires and policy figures urged new regulation in response.
Why the threat has real teeth
Nvidia's leverage extends beyond chip supply. The company has committed roughly $40 billion to equity deals tied to major AI companies, including OpenAI and Anthropic, according to a Startup Fortune analysis of its investment activity — a strategy that secures demand for its chips while deepening the industry's dependency on its technology. Layer financing guarantees on top, and Nvidia's commercial relationships give it both the motive and the means to monitor how its customers behave.
That combination — dominant market share, capital entanglement and now an explicit willingness to withhold supply — makes Huang's hypothetical something closer to a governance mechanism. As Startup Fortune put it, when the company that makes nearly everyone's chips says it would refuse to sell to someone, that is "a company admitting it has a kill switch it has never used."
Critics will note the tension in Huang's position: he argues against new AI regulation while describing a private company wielding quasi-regulatory power over the entire industry — power that is accountable to shareholders rather than the public. Supporters counter that a supplier with skin in the game is more incentivized to act responsibly than a distant bureaucracy.
The statement also lands at a crowded moment for AI-safety news. On the same day, chip-equipment maker ASML raised its guidance for the third time on surging AI demand, per the Wall Street Journal — evidence that the industry Huang oversees is still in a furious buildout phase with no off-ramp in sight. Nvidia has also begun organizing in Washington, launching an employee-funded political action committee to engage on AI policy, a step that suggests the company expects more scrutiny rather than less.
What a cutoff would actually look like is far from clear. Supply agreements for AI chips run to billions of dollars and are negotiated years in advance; unilaterally halting shipments would invite litigation, accelerate customers' bets on rival accelerators, and test Nvidia's own revenue guidance. Huang's point may be precisely that the deterrent should never need to be used — a public statement of willingness, backed by market dominance, designed to make every lab think harder about the systems it ships.
Either way, the statement sets a marker for the AI era: the world's most important chipmaker says it would rather lose revenue than be the arms supplier to a catastrophe. Whether that promise would survive its first real test — a paying customer, hundreds of millions of dollars in orders, and a safety judgment made in-house — is a question Huang, for now, has only answered in the hypothetical.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
