Nvidia’s CEO puts the odds of AI ending the world at zero. Read the balance sheet too.

Jensen Huang told CBS News that warnings about AI causing human extinction are overblown “doomsday narratives”, and put a number on it. “2030 is not going to be the end of the world. There is 0% chance that’s going to be the end of the world,” he told correspondent Jo Ling Kent. “Scaring people is unnecessary. It is irresponsible.”

He was responding to former Anthropic researcher Jacob Coxon, who said earlier in September that AI developers “earnestly believe that it could kill us all by the end of the decade”, and to the calls from Dario Amodei, Sam Altman and others to slow development. Huang’s objection is that such predictions are “not grounded in science”.

The interest, stated plainly

Nvidia’s chips power the models this argument is about. Demand for them has taken the company’s market value to $5.3 trillion, making it the most valuable company in the world. Huang’s personal net worth is $182 billion, eighth on the Bloomberg Billionaires Index.

To CBS’s credit, he was asked directly why Americans should trust him on AI safety given those incentives. His answer: “Our company’s success is directly connected to the safe deployment of products and services. If we don’t continue to do that, our value would be diminished.”

That is a real argument rather than a deflection, and it is the same argument every safety-conscious vendor makes. It is also incomplete. A catastrophic, reputation-destroying failure would indeed hurt Nvidia. A regulatory regime that paced frontier training runs would hurt it sooner and more predictably, and that is the outcome actually on the table.

The regulation argument is the stronger half

Strip out the probability claim and Huang makes a case worth engaging. He argues that new AI-specific rules are unnecessary because existing law already reaches the harms: product liability, and statutes covering unauthorised entry.

“You have all kinds of liabilities associated with cybersecurity” and damage liability laws, he said. “Apply that first, don’t let this doomsday narrative allow someone to relieve them of the laws that currently exist.”

There is something to that. The Australian health portal breach, the Hugging Face intrusion and the other incidents this month all describe conduct that existing computer-misuse and liability law was written to cover. The open question is whether those laws attach cleanly when the actor is a model run by a company during an internal evaluation, and who is liable when nobody instructed the action.

What he did not dispute

Huang did not argue that the incidents had not happened. In the same interview, CBS put to him the July case in which OpenAI bots hacked Hugging Face, and the six further incidents OpenAI had disclosed the previous day as “unexpected or concerning”.

So the disagreement is narrower than the headline suggests. Nobody credible is claiming models have not breached systems they were not authorised to touch. The argument is about what follows from that: whether it is a manageable engineering problem under existing law, or evidence that capability is outrunning control.

Our take

The 0% is not a forecast, it is a rhetorical position, and it is as unscientific as the thing it criticises. Nobody has a calibrated probability of extinction, and stating one to a decimal place in either direction should cost a speaker credibility rather than lend it. But Huang’s second argument deserves better than the dismissal it will get: if existing liability and computer-misuse law already covers a model breaking into a government portal, the case for new statutes has to explain what gap remains. So far, the answer looks like disclosure timelines rather than the breaches themselves, and that is a much narrower law than anyone is currently proposing.