Nvidia CEO Jensen Huang warned that China has the massive computing power and infrastructure needed to match the capabilities of Anthropic’s high-level AI model, Claude Mythos.
Summary
- Nvidia CEO Jensen Huang warned that China’s massive infrastructure and “ghost databases” provide the computing power needed to match the capabilities of the Anthropic Claude Mythos.
- Huang said the hardware used to train the revolutionary model is largely available in China and suggested the Chinese government could easily integrate the chips to counter American AI advances.
Speaking on Dwarkesh Patel’s podcast on Wednesday, Huang dismissed the notion that China lacks the hardware to match Western AI success.
While Washington views the recently released Mythos model as an important asset, Huang noted that the capabilities needed to train such a system are “largely available” in China.
He pointed to a large area of ”datacenters” that remain empty but fully powered, suggesting that the Chinese government could “add more chips” at any time to counter American advances.
The urgency of this measurement comes from the specific cyber attack capabilities found within the Anthropic model. Internal testing in April revealed that Mythos could identify thousands of software vulnerabilities in browsers and apps, prompting the company to limit access to prevent abuse. The AI Security Institute supported these concerns on April 13, finding a model capable of independent, multi-level attacks on networks.
Financial institutions remain particularly vulnerable to these advanced AI hackers because they rely on decades-old software, according to a Reuters report published on Tuesday.
This fear is based on recent history; Anthropic reported in November that state-sponsored actors from China had already attempted to use the “Claude Code” tool to infiltrate a number of global targets.
Balancing competition and discussion
US Treasury Secretary Scott Bessent described Mythos as a “step change” in learning skills that can protect America’s leadership.
Huang, however, argued that treating China as a target for restrictions could be dangerous, as the country produces 60% of the world’s conventional chips and hosts half of all AI researchers.
“Harassing them, making them an enemy, is probably not the best answer,” Huang said.
Although he has identified China as an enemy and expressed a desire for the United States to win the technology race, he maintains that the size of China’s infrastructure requires a more flexible approach.
“I think having a conversation and having a research discussion is probably the safest thing you can do,” he added, suggesting that open communication may be the only way to reduce the global risks posed by such powerful autonomous tools.
Last week, Bessent met with executives at America’s central bank last week to address what he described as “unprecedented” cybersecurity risks posed by the next generation of autonomous AI.
