US approved Nvidia H200 chip sales to Chinese firm with empty office

The United States approved Nvidia H200 AI chip sales to New Matrix, a Chinese company whose listed office appeared to be empty.

Reporters for The New York Times found an empty gray area at the listed office, with ceiling wires dangling where lights should have been installed, and little evidence that the company actually operates there. The company’s name did not appear in the building directory at its listed address. Real estate management agency staff at the industrial park said they had no files on the company. The company came into existence 17 months ago, and its owners are tied to a Chinese private investment company placed on a US government blocklist in 2024.

Other companies tied to organizations under national-security trade limits also received permission to buy chips. While Joe Biden was president, Nvidia faced limits on chip sales to China. The Donald Trump administration eased those restrictions after Nvidia agreed to hand over 15% of all sales to the government. CEO Jensen Huang ran an active lobbying campaign.

Since 2022, the US has restricted sales of Nvidia’s most advanced chips to China through export controls tied to national security concerns. Washington first required licenses for sales of Nvidia’s A100 and H100 AI chips to China in August 2022. Nvidia answered by creating weaker versions for the Chinese market, including the H20.

In April 2025, the US restricted those chips too, including the H20 and other weaker versions Nvidia had designed for the Chinese market, and Nvidia recorded a $4.5 billion charge for chips it could no longer sell. Before the tighter rules, Nvidia held nearly 95% of China’s advanced AI chip market, and China represented roughly 13% of company revenue.

Under a new rule, cloud providers buying Nvidia chips must direct at least 51% of services to companies outside China, capping Chinese customers at 49% of business. It covers new contracts and applies to Nvidia partners worldwide, not just in Asia.

On Oct. 3, the Financial Times reported that Nvidia instructed data center clients to reduce work for Chinese customers. The company delivered the instruction verbally through sales and regional teams instead of in writing. Nvidia did not confirm or deny the report.

Nvidia told AFP that “data centers built on American chips are the foundation U.S. AI developers need to compete globally.” Mao Ning, a spokeswoman for China’s foreign ministry, said businesses from every country, including Chinese firms, deserve a fair, transparent and non-discriminatory setting.

The recent Donald Trump-Xi Jinping summit did not cover export controls, and neither side appeared interested in making concessions. Analysts said a two-month extension of the trade truce could end once new US controls arrive. Analysts expect looming US controls could prompt China to retaliate by halting rare earth exports.

Chinese officials are reviewing their own restrictions on advanced chip imports. China is considering relaxing controls so some of its largest AI firms can import perhaps millions more of Nvidia’s banned chips over the next year. China’s Ministry of Industry and Information Technology asked Alibaba and ByteDance to provide plans for buying Nvidia RTX Pro 5500 chips. Beijing asked the companies to explain how they would use the gaming chips, expecting servers could run them to support China’s highest-demand AI models.

Alibaba and ByteDance are among Chinese companies that have rented chips rather than buying them, using overseas data centers in Singapore and Malaysia that non-Chinese companies own to train AI models. Yahoo! Finance reported that this path could become harder under the new rule.

Donald Trump looks to Nvidia Chief Executive Officer Jensen Huang for AI advice, although not everyone believes Nvidia’s interests align with those of the United States. Some critics argue Nvidia’s profit motives could shape Trump’s views at a crucial moment when a president’s understanding of emerging technology risks may matter.