PLA records show repeated bids for restricted Nvidia chips, analysis finds

Procurement records from the People’s Liberation Army show a sustained effort to obtain advanced Nvidia chips that Washington has tried to keep out of China, raising a blunt enforcement question: are export controls deterring access, or simply forcing it into more creative channels?

An analysis of about 3,800 Chinese military procurement records found more than 500 instances since 2019 in which PLA units sought Nvidia chips by name or technical specification, including the restricted A100 and A800 graphics processors. Those chips have been subject to U.S. export limits since 2022, when Washington moved to curb sales of advanced AI hardware over concerns that it could strengthen China’s military and artificial intelligence capabilities.

The records suggest the demand is not limited to one project or one unit. In 2025, PLA procurement documents referenced Nvidia hardware for AI servers, systems linked to DeepSeek, and even a 33-pound robot dog. One request for an Nvidia Jetson module for the robot was later canceled, but the paperwork still showed how widely military-linked buyers were trying to spread restricted American technology across computing and robotics projects.

The pattern has sharpened concerns among U.S. officials and lawmakers about how easily controls can be bypassed through intermediaries and overseas procurement routes. Public tender documents reviewed this year showed four Chinese universities, including two with PLA ties, buying Super Micro Computer servers containing restricted AI chips over the past year, while two other military-linked institutions sought similar purchases. In Washington, two senators pressed Commerce Secretary Howard Lutnick to consider pausing export licenses for advanced Nvidia AI chips and server systems headed to China or routed through intermediaries in Southeast Asia.

The issue has become especially pressing as the Commerce Department moves to close loopholes that could let advanced chips reach Chinese entities outside China through overseas subsidiaries. That action came after U.S. officials in May cleared around 10 Chinese firms to buy Nvidia’s H200 chip, though no deliveries had been made. The mixed signals underline the policy tension: the United States is still licensing some high-end sales while trying to seal off the routes that make those sales meaningful.

Nvidia has argued that China already has “more than enough” domestic chips for military use and that restricted products would be a “nonstarter” for military applications without support, software or maintenance. The procurement records suggest the larger problem is not just the chip itself, but the network of buyers, brokers and shell entities built to keep advanced U.S. technology flowing despite the rules.