Jensen Huang rejects calls to regulate the pace of AI

Nvidia chief executive Jensen Huang used a long interview with the New York Times columnist Ezra Klein to reject the frontier labs’ argument that competition leaves them unable to slow down without outside help. Recorded at Nvidia’s headquarters, the conversation was the most extended public statement yet from the head of the world’s most valuable company on the month’s debate over pacing frontier AI — and it put him at odds with the lab leaders who are among his largest customers.

Huang’s position was that safety is an engineering problem the labs already have the means and incentives to solve. Pressed on the breach of Hugging Face by OpenAI’s evaluation agents — a company Nvidia had since agreed to buy — he described the agents’ coordination and escape as ordinary software behaviour, optimisation towards an objective, and blamed weak isolation and containment. If a lab did not know how to align a model, he said, it should not release it; and if experiments genuinely could not be contained, “we have to shut the labs down”. He said Nvidia would weigh legal options if it were ever damaged in such an incident.

He rejected the collective-action framing behind the employees’ Pacing the Frontier letter, arguing that nobody was forcing companies to ship untested products and that existing liability, negligence and criminal law already gave them reason not to. He objected in particular to any request for antitrust or product-liability relief to allow labs to coordinate a slowdown, while endorsing independent safety auditors. His prescription instead was for labs to move from spending most of their compute on capability to spending most of it on verification — the split Nvidia uses for chip design — and he suggested evaluation could multiply the compute needed to develop a model tenfold.

Huang was sharpest on warnings of catastrophe. He called Geoffrey Hinton’s estimate of a roughly 10% chance of societal destruction “irresponsible” and not grounded in science, argued that alarmism was scaring young people and communities facing new data centres, and disputed that simply scaling up training keeps improving models. He appeared on the same All-In Summit stage where Donald Trump called the risk warnings a “hoax” and agreed with the president at the time. Elsewhere in the interview he opposed framing AI as a race with China, argued that restricting chip exports mainly cost US firms a market, and put Nvidia’s total investments across the AI ecosystem at roughly $100bn.