Nvidia Corp. Chief Executive Officer Jensen Huang on has criticized the U.S. government’s restrictions on chip sales to China.

Huang made the remarks at the Computex conference in Taipei Tuesday. During the event, Nvidia Corp. debuted a technology that makes it possible to move data between third-party chips using its NVLink interconnect. The company also updated its Isaac GR00T foundation model for powering robots.

Huang called the U.S. export controls on AI chips a “failure” at Computex. “Chinese AI researchers will use their own chips,” the Financial Times quoted the executive as saying. “They will use the second-best. Local companies are very determined, and export controls gave them the spirit, and government support accelerated their development. Our competition is intense in China.”

The U.S. government first restricted the sale of Nvidia’s most advanced chips to China in 2022. The following year, officials expanded the curbs to cover more of the company’s graphics cards. Exports of certain chipmaking equipment were restricted as well.

This past January, the Biden administration introduced another set of curbs in a document called the Interim Final Rule on Artificial Intelligence Diffusion. It was designed to prevent China from obtaining AI chips via third countries. Under the policy, companies in most nations had to obtain a license to buy processors with “computational power equivalent to up to 320,000 advanced GPUs.”

“The fundamental assumptions that led to the AI diffusion rule in the beginning have proven to be fundamentally flawed,” Huang said. “If the U.S. wants to stay ahead, we need to maximize and accelerate our diffusion, not limit it.”

The Trump administration rolled back the Interim Final Rule on Artificial Intelligence Diffusion earlier this month. A few weeks earlier, officials banned Nvidia from selling its H20 graphics card to China. The H20 is a scaled-down version of the company’s H100 chip that was designed to comply with U.S. export rules.

Huang criticized the ban at Computex and revealed that the company’s share of the AI chip market in China has declined significantly over the past few years. “Four years ago, Nvidia had 95% market share in China. Today, it is only 50%,” he detailed. “The rest is Chinese technology. They have a lot of local technology they would use if they didn’t have Nvidia.”

Huang added that Nvidia’s next chip for the Chinese market won’t be based on the Hopper architecture, which powers the H20 and the H100. The reason is that the company has already “degraded the chip so severely” to comply with export controls, he explained. “It’s not Hopper because it’s not possible to modify Hopper any more.”

After the U.S. banned the sale of H20 chips to China, Nvidia revealed that it would have to make a $5.5 billion inventory writedown. Reuters reported in April that the company had received $18 billion worth of H20 orders during the first quarter. Customers in China accounted for 13% of Nvidia’s revenue, or $17 billion, in the fiscal year ended Jan. 26.

Photo: Nvidia