US Export Approvals for Nvidia H200 GPUs Include Shell Integrators with Military Ties
Department of Commerce export licenses for Nvidia's H200 AI processors have been granted to Chinese server assemblers operating out of empty offices and linked to sanctioned military suppliers.
The U.S. Department of Commerce has issued export licenses allowing Chinese firms to receive Nvidia H200 AI GPUs, but investigation into the supply chain reveals that several approved hardware integrators operate out of vacant facilities while maintaining direct ties to sanctioned military suppliers. Rather than acting as direct enterprise end users, entities such as New Matrix and Maginfra were cleared to purchase high-performance compute silicon to assemble the chips into complete server nodes before final delivery to undisclosed customers.
Physical inspections of registered addresses for approved licensees revealed deserted office spaces devoid of operational infrastructure. According to reporting from The New York Times, cited by both Tom's Hardware and The Verge, the registered office for New Matrix consisted of an empty shell with dangling ceiling wires and no listed presence in building directories. Despite having no online operating history, public records show New Matrix was established 17 months ago by executives connected to Wise Road, a private equity firm sanctioned by Washington for acquiring sensitive semiconductor technologies for Beijing.
Ownership structures for these intermediate hardware assemblers trace back to broader state-backed telecom and military equipment operations. New Matrix founder Li Bin chairs state-linked conglomerate Tsinghua Unigroup and owns New H3C Semiconductor Technologies, an entity blacklisted by the Commerce Department for military cooperation. Similarly, Maginfra shares personnel, patents, and government ownership stakes with Inspur, a server manufacturer placed on the U.S. Entity List in 2023 for supplying the People's Liberation Army.
The regulatory approvals also extend to third-party infrastructure providers supporting military suppliers. Licensee Zhejiang Yindun Cloud Technology maintains executive linkages to the founder of Range Intelligent Computing Technology Group, a data center operator providing compute capacity to defense contractors. Zhejiang Yindun reportedly operates part of its deployment inside Range's data center facilities, raising technical questions about how hardware utilization and physical access controls are monitored post-delivery.
Export restrictions on Nvidia's high-bandwidth memory and accelerator architecture were previously tightened under export controls, but sales resumed following policy shifts under the Trump administration. The Verge reports that sales were allowed after Nvidia agreed to give the government a 15-percent cut, whereas Tom's Hardware reports the agreement involved a 25-percent cut of sale proceeds. While established tech conglomerates including Baidu, Alibaba, Tencent, ByteDance, and Byte-backed platforms received licenses, former National Security Council official Liza Tobin warned that current licensing protocols lack technical due diligence to verify post-assembly hardware destinations.
Sources
- 01 Two Chinese companies cleared to buy banned Nvidia AI chips have empty offices and unknown addresses — Tom's Hardware
- 02 The US let Nvidia sell chips to an empty office in China — The Verge — AI