Nvidia CEO Jensen Huang sparked a heated debate during a recent podcast appearance with Dwarkesh Patel, where the two discussed whether the United States should continue selling advanced AI chips to China.
During the conversation, Patel questioned whether giving China access to high-powered AI hardware is ultimately beneficial. Known for playing devil’s advocate in interviews, he challenged Huang directly, asking whether such sales could pose a threat to American companies and even national security.
To support his concerns, Patel cited Anthropic’s Claude Mythos, which he claimed uncovered “thousands of zero-day vulnerabilities” across major operating systems and web browsers. He argued that if China had access to the immense computing power Nvidia provides, it could potentially use it to develop advanced cyber-offensive capabilities that might endanger U.S. security.
Huang responded with a more measured take, pushing back on the severity of the claim. He noted that Mythos was trained using what he described as “fairly mundane capacity” and a relatively modest amount of computing power, suggesting that the risks may be overstated.
