China Has Compute to Build Its Own Mythos AI: Huang

Jensen Huang just made the chip export control argument a lot harder to sustain. The Nvidia CEO said this week that China already has the computing power to train an AI model comparable to Claude Mythos — Anthropic's new system that was quietly locked down after it started finding zero-day vulnerabilities at scale. If Huang is right, the entire premise of restricting Beijing's access to high-end chips has a significant hole in it.

What Did Jensen Huang Actually Say?

Speaking on the Jensen Huang Dwarkesh Patel podcast on Wednesday, Huang was asked a direct question: could China train a model like Claude Mythos — one with demonstrated cyberoffensive capabilities — and would that be a threat to US national security? His answer was blunt. The compute needed for Mythos was trained on 'fairly mundane capacity,' he said, and that capacity is 'abundantly available in China.'

Huang didn't hedge. China has 'enormous' compute, he said, pointing to a string of structural advantages that Washington's export restrictions were supposed to close off. The country manufactures 60% of the world's mainstream chips. It has some of the best computer scientists on the planet. And crucially — 50% of the world's AI researchers work in or for Chinese institutions. Energy, too, is not the bottleneck.

This is the part of the interview that deserves serious attention. The chip export controls that the US has pushed since 2022 were built on a theory: cut off access to cutting-edge GPUs, and you cut off access to frontier AI. Huang is saying, at minimum, that theory doesn't apply to Mythos-class training runs.

Victimizing them, turning them into an enemy, likely isn't the best answer. They are an adversary.

Why Claude Mythos Makes This Conversation Urgent

Anthropic released findings on Claude Mythos Preview on April 7, and the reaction from the security community was swift. The model identified thousands of software vulnerabilities across major operating systems and browsers — and the company disclosed that 99% of those vulnerabilities have not been patched yet. Anthropic moved to limit access shortly after. That's not a normal product decision. You don't gate a commercial AI model unless the threat surface is real.

On April 13, the AI Security Institute published its own evaluation. The findings were stark: Mythos can execute multi-stage attacks on vulnerable networks and discover and exploit vulnerabilities autonomously — work that would take trained human professionals days. Banks running decades-old software are particularly exposed, according to reporting from Reuters this week.

That's the context behind Huang's warning. A Chinese-built model with Mythos-level capabilities, deployed without Anthropic's access restrictions, wouldn't just be a national security concern in theory. The offensive surface would be massive.

Washington Disagrees — But Is Anyone Listening to Huang?

On Tuesday — one day before Huang's interview dropped — US Treasury Secretary Scott Bessent praised Mythos as the thing that would keep America ahead of China in the AI race. 'This Anthropic Mythos model was a step function change in abilities, learning capabilities,' Bessent told Bloomberg. He framed it as proof that American AI leadership is intact.

Huang's comments land differently. He's not arguing that the US isn't ahead — he's saying China has the raw infrastructure to close the gap on this particular capability, regardless of what chips it can officially import. Those are two very different claims, and Washington appears to be focused only on the first one.

Last November, Anthropic itself reported that a Chinese state-sponsored group had manipulated its Claude Code tool in an attempt to infiltrate roughly 30 global targets, succeeding in a small number of cases. That operation predates Mythos. The escalation potential from here is not abstract.

Call it a frank assessment from a man whose company has a stake in selling chips everywhere — but Huang's read on Chinese compute capacity is probably more accurate than most government briefings. He's in the supply chain. He knows what's been built.

This Anthropic Mythos model was a step function change in abilities, learning capabilities.