NVIDIA CEO Jensen Huang has warned that China already possesses the computing infrastructure and semiconductor capacity necessary to train an artificial intelligence model at the same level as Anthropic’s recently released Claude Mythos, a system that has raised global cybersecurity concerns since its April debut.
Speaking on the Dwarkesh Patel podcast on Wednesday, Huang pushed back against the notion that U.S. export controls have materially limited China’s ability to develop advanced AI systems, stating that the hardware used to train Claude Mythos is already widely available across Chinese data centers.
China’s Computing Capacity Extends Beyond Export Restrictions
“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,” Huang said during the interview. He noted that the model was trained on what he described as “fairly mundane capacity,” suggesting that China could replicate similar workloads using existing infrastructure.
Huang highlighted several factors that reinforce China’s position: the country manufactures approximately 60% of the world’s mainstream semiconductor chips, hosts roughly half of all global AI researchers, and maintains abundant energy reserves.
He also pointed to underutilized data centers across the country, describing them as “ghost data centers” that are fully powered and operationally ready but remain largely idle.
“They have ghost cities, they have ghost data centers too,” Huang remarked, adding that even with access limited to 7-nanometer chip processes, China could compensate by deploying larger numbers of processors.
Huang Calls for US-China AI Safety Dialogue
While Huang identified China as an adversary and expressed a clear desire for the United States to maintain its technological lead, he argued that a purely restrictive approach could prove counterproductive. Instead, he called for open dialogue between U.S. and Chinese AI researchers to establish shared guidelines on how the technology should be used and should not be used.
“We want the United States to win, but I think having a dialogue and having a research dialogue is probably the safest thing to do,” Huang said, adding that it is “essential” for researchers on both sides to be in communication.
The comments carry particular weight given Anthropic’s decision to restrict access to Claude Mythos after the model identified thousands of software vulnerabilities across major operating systems and browsers. The finding has intensified concerns that similar capabilities, if developed without safeguards, could be weaponized for large-scale cyberattacks.
NVIDIA Deepens Its Stake in AI Development
Huang’s remarks come as Nvidia continues to expand its presence in the AI sector. The company has reportedly committed $10 billion in investment to Anthropic, the maker of Claude Mythos, though Huang has suggested this may represent Nvidia’s final major investment in the company.
According to Bloomberg, Huang framed the Mythos breakthrough as evidence that U.S.-China cooperation on AI safety is no longer optional but necessary for global stability.
