A new investigation has raised serious concerns about the effectiveness of American semiconductor export controls after the United States reportedly approved NVIDIA H200 AI chip transactions involving Chinese companies with empty offices and connections to sanctioned businesses. According to a November 7 investigation by The New York Times, the US Commerce Department cleared approximately 2 dozen Chinese companies to purchase or handle NVIDIA's advanced AI accelerators, including businesses whose ownership structures raise questions about potential connections to China's military technology ecosystem.
One of the companies, New Matrix, received authorization to assemble NVIDIA H200 accelerators into AI servers despite its registered Beijing office appearing completely vacant. Investigators found an unfinished commercial space with exposed wiring, no company directory listing, and building management unable to confirm the company's presence. Founded approximately 17 months earlier, New Matrix reportedly has ownership connections to Wise Road Capital, a Chinese investment firm placed on the US Entity List in 2024 for activities involving strategically sensitive semiconductor technologies. Its founder, Li Bin, also chairs Tsinghua Unigroup, whose subsidiary New H3C Semiconductor Technologies was previously restricted over reported cooperation with China's military.
Another approved company, Maginfra, presented similar concerns. Investigators visiting its listed office discovered an apparently unused room containing only a few chairs and desks. Analysts believe Maginfra may be connected to Inspur, a major Chinese server manufacturer sanctioned by Washington in 2023 over military related activities. Both companies reportedly share personnel, patents, and government ownership connections. Importantly, New Matrix and Maginfra were authorized to assemble AI servers rather than necessarily purchase H200 accelerators for their own computing operations.
The investigation also identified other approved companies with reported links to military suppliers. Zhejiang Yindun Cloud Technology maintains connections to Range Intelligent Computing, which provides cloud infrastructure for projects associated with Chinese defense organizations. Meanwhile, Shanghai Jiatang originated from Datang Telecom, a state controlled telecommunications company involved in military communications technology. These corporate relationships raise concerns about potential indirect access to restricted computing hardware, although the investigation does not establish that NVIDIA H200 chips were delivered to the Chinese military or that the approved companies violated export regulations.
The findings follow the Trump administration's decision to relax restrictions on certain AI semiconductor exports in December 2025. Under the US Commerce Department's January 2026 licensing rules, NVIDIA H200 and AMD MI325X exports can be reviewed individually, provided applicants meet security, customer verification, and supply requirements. However, the discovery of questionable corporate addresses and ownership connections suggests that verifying the final destination of advanced AI hardware remains difficult. The controversy also follows Jensen Huang's position on maintaining American semiconductor competition in China, while keeping NVIDIA's most advanced Blackwell and Rubin processors restricted.
Despite these approvals, direct H200 sales to China remain relatively limited. Shipments to Chinese customers accounted for less than 1% of NVIDIA's data center revenue during the quarter ending in July 2026. The latest findings could nevertheless intensify scrutiny of export licensing procedures and the responsibilities of semiconductor manufacturers when verifying international customers.
The investigation exposes a potential weakness in semiconductor export regulations: restricting advanced hardware accomplishes little if authorities cannot reliably verify the companies authorized to purchase, assemble, or redistribute it.
However, corporate connections to sanctioned businesses should not automatically be treated as proof of illegal activity. Effective enforcement requires transparent ownership verification, proper investigation, and evidence of actual violations.
NVIDIA and American regulators face the challenge of balancing international semiconductor competition with national security. Greater accountability throughout the supply chain is important, but additional restrictions should target demonstrated risks rather than unnecessarily limiting legitimate technology businesses or consumer access to computing hardware.
Should NVIDIA and US regulators introduce stricter verification for companies purchasing advanced AI accelerators, or are existing export restrictions already creating unnecessary barriers to international semiconductor trade?
