Nvidia CEO Jensen Huang warned that China already possesses the computing power and data-center capacity required to train an AI model comparable to Anthropic’s Claude Mythos, a capability he said could threaten global cybersecurity.
In an interview on the Dwarkesh Patel podcast, Huang said the compute used to train Mythos was “fairly mundane” and that the necessary chips and infrastructure are abundantly available in China. “The amount of capacity and the type of compute it was trained on is abundantly available in China, so you just have to first realize that chips exist in China,” he said.
Anthropic restricted access to Mythos in April after the model identified thousands of software vulnerabilities across major operating systems and browsers, raising concerns about potential misuse in cyberattacks. A Chinese-made model with similar capabilities, Huang warned, “could wreak havoc if misused.”
Huang emphasized the scale of China’s resources: he noted the country manufactures roughly 60% of the world’s mainstream chips, employs top computer scientists, accounts for about half of global AI researchers, and has abundant energy. He described large, sometimes idle capacity: “They have datacenters that are sitting completely empty, fully powered. 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.”
Despite framing China as an adversary, Huang urged engagement over confrontation. “Victimizing them, turning them into an enemy, likely isn’t the best answer,” he 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.”
U.S. officials have highlighted Mythos’s significance. On Tuesday, U.S. Treasury Secretary Scott Bessent described Mythos as a “step function change in abilities, learning capabilities,” saying it helps keep America ahead in the AI race.
Anthropic’s disclosure about Mythos sparked broader security assessments. The company said 99% of the vulnerabilities the model found remain unpatched. The AI Security Institute (AISI) evaluated Mythos and concluded it could “execute multi-stage attacks on vulnerable networks and discover and exploit vulnerabilities autonomously,” performing tasks that would take human experts days.
Security analysts and reporters have warned that AI-augmented attacks could threaten institutions such as banks, which often run legacy software, and other critical systems. Anthropic also reported last year that a “Chinese state-sponsored group” attempted to misuse its Claude Code tool to target about 30 global entities, succeeding in a small number of cases.
The conversation underscores tensions between the defensive risks posed by advanced AI models and the geopolitical realities of distributed compute and talent. Huang’s remarks call attention to how widespread compute capacity and research capability could enable similar models to be developed outside the companies that created them, with implications for cybersecurity and international policy.
Cointelegraph is committed to independent, transparent journalism. This article is produced in accordance with Cointelegraph’s Editorial Policy; readers are encouraged to verify information independently.