US Permits Import of Downgraded AI Chips to China

The Trump administration has shifted its stance on the import of AI chips to China, allowing Nvidia to sell processors as the company “does not transfer the best technology,” stated Commerce Secretary Howard Lutnick in an interview with CNBC.

According to him, the chipmaker plans to sell China the “fourth-tier” chip, which is slower than the most advanced solutions used in the US.

“We are not selling them the best, second best, or even third best. Only the fourth,” he said.

In July, Nvidia announced its intention to resume sales of H20 AI chips in China. AMD made a similar statement.

Lutnick added that the resumption of supplies is linked to a rare earth metals deal. He noted that it is in the US interest for Chinese companies to continue using American developments, maintaining a connection to the “technology stack.”

Nvidia CEO Jensen Huang advocated for the continued sale of chips to China, warning that otherwise local companies might start investing in domestic infrastructure. Previously, he downplayed US concerns about the use of the company’s chips in China’s military sector.

“The idea is that the Chinese are more than capable of creating their own system. We just need to maintain technological superiority by at least one step so they continue buying our chips,” noted Lutnick.

The US administration changed its decision after President Donald Trump met with Huang in Washington.

“We need to sell the Chinese enough so that their developers get hooked on the American technology stack. That’s the logic,” added the Commerce Secretary.

The H20 was introduced as a response to export restrictions by the Biden administration. It is based on the same architecture as the Hopper generation processors, which are supplied in the US as part of complete systems with H100 or H200 chips.

For the sale of H20 in China, Nvidia reduced some of its features—cutting the number of GPU cores and decreasing the bandwidth between chip components.

The current ranking of the company’s top AI products is as follows:

• Blackwell GB200 — includes both graphics and central processors in one module;
• Blackwell B100 and B200 — simplified versions;
• there are also Blackwell chips for gaming and graphics, which can be used for artificial intelligence but are weaker than specialized solutions.

In March, Nvidia announced new processors for creating and running AI models—Blackwell Ultra and Vera Rubin.

Nvidia’s Collaboration with China

Huang praised China’s achievements in open-source artificial intelligence and promised collaboration with local companies.

He stated that large language models from Chinese firms like DeepSeek, Alibaba, Tencent, MiniMax, and Baidu are on par with the best global developments. They “were created here, openly distributed, and have contributed to the advancement of technology worldwide.”

Chinese open-source developments have become a “catalyst for global progress,” providing all countries and industries with a chance to join the AI revolution, added the top manager.

Huang spoke at the opening ceremony of the China International Supply Chain Achievements Exhibition.

“I am very pleased that the export restrictions on H20 have been lifted, and now we can serve this market,” he noted.

Nvidia announced plans to release a new, fully compliant RTX PRO graphics processor for the Chinese market. It will be “optimal for use in AI solutions for smart factories and logistics.”

Huang also emphasized his firm’s role in the achievements of Chinese tech companies. He stated that China has achieved “extremely rapid technological progress” thanks to the efforts of its researchers, developers, and entrepreneurs, and today more than 1.5 million engineers in the country are creating solutions based on Nvidia platforms.

In July, the Alibaba-backed Chinese AI startup Moonshot AI released a new open-source model, Kimi K2.