China’s military is seeking advanced Nvidia artificial intelligence chips for military applications despite U.S. export restrictions, according to a congressional commission report released on Nov. 18.
The U.S.-China Economic and Security Review Commission (USCC) cited research based on Chinese military procurement records showing requests for American processors to support AI research, including work related to military sensors and computer simulations.
The findings come as China’s People’s Liberation Army (PLA) increasingly incorporates AI into weapons systems, targeting, battlefield decision-making, and other military operations.
According to the commission, recent PLA exercises have involved AI-assisted target recognition, allocation of firepower, and systems designed to help commanders make decisions.
Military Procurement Records
Researchers at Georgetown University’s Center for Security and Emerging Technology examined thousands of PLA procurement documents and identified requests for advanced American chips, including models subject to U.S. export controls.
In their analysis, researchers Cole McFaul and Sam Bresnick described a May 2025 request from a PLA research laboratory seeking 16 Nvidia H100 processors for AI modeling involving electromagnetic and infrared characteristics—research relevant to military sensing and detection.
Another military request, issued in July 2025, sought a server equipped with Nvidia H20 processors for computer simulations that the researchers said could include weapons-system modeling.
Separate Chinese military procurement records also document demand for advanced Nvidia hardware.
A July 2024 notice published on China’s official Military Procurement Network specified four Nvidia H100 processors for a server intended to perform AI calculations and run large AI models.
The documents establish that Chinese military buyers sought the technology, but they do not state whether the chips were purchased or delivered. The Georgetown researchers said the available records did not allow them to determine the overall scale of the PLA’s use of U.S. processors.
China Still Trails in Advanced Chips
The commission also identified technological obstacles to Beijing’s effort to replace American AI processors with Chinese alternatives.
In a separate chapter, the commission compared Huawei’s Ascend 910C processor with Nvidia’s more powerful B200, finding that the U.S. chip offered approximately three times the processing performance and 2.5 times the memory bandwidth—the speed at which data moves to and from the chip.
The report said Chinese manufacturers also face difficulties producing advanced chips because they lack access to some of the most sophisticated manufacturing equipment.
As a result, Huawei needs more chips, electricity, and supporting equipment to achieve computing performance comparable to that of leading U.S. processors, according to the commission.
US Moves to Restrict Remote Computing Access
The Georgetown researchers also found that Chinese military buyers were seeking access to U.S. computing technology without necessarily purchasing the chips themselves.
A July 2024 PLA procurement request sought 10 million core-hours of supercomputer processing time, specifying that an outside provider must be able to supply computing resources using Nvidia V100 or A100 processors. A core-hour measures the use of one processor core for one hour.
Such arrangements could allow military customers to use U.S. computing technology through outside providers rather than obtain the physical processors.
On Nov. 16, the Commerce Department’s Bureau of Industry and Security announced new restrictions aimed at preventing sensitive American computing technology from reaching restricted users through data centers in third countries.
The rule, scheduled for publication in the Federal Register on Nov. 19 and set to take effect upon publication, establishes additional licensing requirements for certain advanced chips and overseas data centers providing substantial computing power remotely to companies based in China and other restricted destinations.
Data centers already providing covered services will have 90 days from the effective date to obtain licenses if they wish to continue.

