The US let Nvidia sell chips to an empty office in China

U.S. export controls permitted Nvidia to sell its banned H200 AI chips to two Chinese companies—New Matrix and Maginfra—that have empty offices and unknown addresses, according to an investigation reported by Tom's Hardware. These firms are linked to executives connected to Wise Road, a Chinese private equity firm sanctioned by the U.S. for helping Beijing acquire sensitive chip technologies, and Tsinghua Unigroup, a tech conglomerate partly owned by the Chinese government. New Matrix's website reportedly lacks information about its owners or history, while its founder chairs Tsinghua Unigroup and owns New H3C Semiconductor Technologies, which is on the Commerce Department's Entity List for cooperating with the Chinese military.

Maginfra is suspected of being linked to Inspur, a company blacklisted by Washington since 2023 for supporting the People's Liberation Army, with both companies sharing personnel, patents, and similar government ownership structures. These two firms are supposed to assemble AI chips into servers for final users, though their customers remain unclear. Another approved company, Zhejiang Yindun Cloud Technology, has executives related to the founder of Range Intelligent Computing Technology Group, a data center operator that has sold compute capacity to Chinese military contractors.

The U.S. Department of Commerce has approved around two dozen Chinese firms to purchase H200 GPUs, including major players like Baidu, BYD, Tesla Shanghai, ZTE, Alibaba, Tencent, and ByteDance, according to The New York Times. Former White House National Security Council China director Liza Tobin expressed concerns that the Commerce Department and Nvidia are not conducting adequate due diligence to ensure licensing terms are followed.

Nvidia CEO Jensen Huang publicly rejected calls for a total ban on AI chip exports to China, stating he "completely disagrees" with Anthropic's Dario Amodei's push for sweeping trade bans. Huang emphasized that American companies must compete globally rather than retreat, noting that Nvidia's next-generation Vera Rubin architecture is already shipping in volume production to domestic research labs and U.S. companies like OpenAI and Anthropic, giving the U.S. several years of lead over China.

The company is also tightening tracking of AI chips in Southeast Asia by requiring buyers to join its fleet intelligence tracking system to become official Nvidia Cloud Partners, a move aimed at preventing billions of dollars worth of restricted hardware from reaching China through smuggling networks. Recent seizures include $12.1 million of advanced AI chips at Malaysia's main airport and the arrest of Super Micro Computer co-founder Wally Liaw on charges tied to an alleged $2.5 billion scheme to route Nvidia GPU servers into China.

Nvidia reportedly cut half of Asian firms on its pre-approved buyer list and told cloud customers that no more than half of their Nvidia computing capacity should be rented to Chinese clients, balancing compliance risks against maintaining demand in one of the fastest-growing AI infrastructure regions.