Jensen Huang doesn’t often make headlines for saying things that undercut the US government’s strategic assumptions about China. But on the Dwarkesh Patel podcast on Wednesday, that’s essentially what he did, and the implications hit hard.
The Nvidia CEO, speaking in a wide-ranging interview, addressed a pointed question about whether China’s access to chips poses a genuine national security threat given the capabilities of Anthropic’s Claude Mythos, a new AI model so powerful in cyberoffensive terms that Anthropic hasn’t released it publicly.
The question touched on one of the most contested debates in Washington right now: whether US chip export controls to China are actually containing anything. Huang’s answer, stripped of diplomatic niceties, was essentially: no, they’re not. Not when it matters.
What Huang actually said and why it matters
Mythos, Huang said, was trained on “fairly mundane capacity,” meaning the computing infrastructure required to produce it isn’t some rare resource that only a handful of countries can access. “The amount of capacity and the type of compute it was trained on is abundantly available in China,” he said. “So you just have to first realize that chips exist in China.”
He didn’t stop there. China manufactures 60 percent of the world’s mainstream chips, is home to some of the world’s best computer scientists, and accounts for roughly 50 percent of AI researchers globally. The country also has an abundance of energy to power compute infrastructure.
Then came the “ghost datacenter” line that’s being widely quoted. “They have datacenters that are sitting completely empty, fully powered,” Huang said. “You know, they have ghost cities, they have ghost datacenters too. They have so much infrastructure capacity. If they wanted to, they could just gang up more chips.”
The argument Huang is that the compute threshold required to get to Mythos-level AI is already within reach. US export controls, in his telling, may be slowing China down on the margins, but they aren’t the barrier that Washington sometimes implies they are.
“Victimizing them, turning them into an enemy, likely isn’t the best answer,” Huang said. “We want the United States to win. But I think having a dialogue and having research dialogue is probably the safest thing to do.” He also said it is “essential that we try to both agree on what not to use the AI for,” a framing that sounds more like arms control negotiations.
Why Claude Mythos made this conversation urgent
To understand why this matters, it helps to understand what Mythos actually is.
Anthropic announced Claude Mythos Preview on April 7 and immediately restricted access to it. The model had autonomously discovered thousands of zero-day vulnerabilities, previously unknown security flaws, across every major operating system and every major web browser.
A zero-day vulnerability is one that has never been publicly disclosed and for which no patch exists yet. The model found one in OpenBSD, an operating system specifically engineered to resist this kind of finding, that had existed for 27 years. Over 99 percent of the vulnerabilities Mythos found have not yet been patched.
Anthropic was clear that these capabilities weren’t intentionally programmed into the model. “They emerged as a downstream consequence of general improvements in code, reasoning, and autonomy,” the company said. The same improvements that make Mythos better at patching vulnerabilities also make it better at exploiting them.
The UK’s AI Security Institute independently evaluated Mythos and found that it could execute multi-stage attacks on vulnerable networks and discover and exploit vulnerabilities autonomously, tasks that would take human security professionals days to complete.
On expert-level capture-the-flag cybersecurity challenges, tasks that no AI model could complete before April 2025, Mythos succeeded 73 percent of the time.
Huang’s point is that a Chinese-built model with equivalent capabilities, trained on infrastructure China already has, could present a genuine threat if deployed without the kind of responsible disclosure framework that Anthropic has put in place around Mythos.
US Treasury Secretary Scott Bessent, by contrast, has framed Mythos as a decisive edge, calling it “a step function change in abilities” that secures American leadership over China. Huang’s podcast comments suggest that edge may be thinner, and more temporary, than Bessent implies.