U.S. chipmaker Nvidia provided technical assistance to China’s DeepSeek that helped the company develop artificial intelligence models later used by the Chinese military, according to a letter from the chairman of a U.S. House of Representatives committee seen by Reuters on Wednesday. The claim highlights growing concern in Washington over the potential military applications of advanced AI technologies developed with U.S. expertise.
Representative John Moolenaar, a Michigan Republican who chairs the House Select Committee on China, wrote to U.S. Commerce Secretary Howard Lutnick that Nvidia’s internal records show the company helped DeepSeek achieve major efficiency gains through an “optimized co-design of algorithms, frameworks, and hardware.” According to the letter, the company reported that DeepSeek’s V3 model required only 2.788 million H800 GPU hours for full training, a fraction of the computing time typically required by U.S. developers for frontier-scale AI models. Reuters reviewed the letter.
GPU hours measure the number of hours AI chips must run to train a model, while frontier-scale models refer to leading-edge AI systems produced by U.S. firms such as OpenAI, Anthropic, and Google. The documents cited in the letter cover Nvidia activities from 2024, a period when there was no public indication that DeepSeek’s technology was being applied by the Chinese military. “Nvidia treated DeepSeek accordingly – as a legitimate commercial partner deserving of standard technical support,” Moolenaar wrote.
Nvidia’s H800 chip, specifically designed for the Chinese market, was sold in China before U.S. export controls on the hardware were implemented in 2023. While the company stated that China already possesses sufficient domestic chips for military applications, concerns persist that technical support and chip sales could indirectly strengthen China’s AI capabilities. The U.S. Commerce Department, the Chinese embassy in Washington, and DeepSeek did not immediately respond to requests for comment.
Earlier this month, the Trump administration approved sales of Nvidia’s more advanced H200 chips to China with restrictions barring sales to entities that support the Chinese military. The decision drew criticism from lawmakers concerned that exporting high-powered AI chips could accelerate Beijing’s military AI programs and erode U.S. technological advantages. Moolenaar warned that even sales to ostensibly civilian users risked violating military end-use restrictions and stressed that rigorous licensing and enforcement measures were necessary to prevent U.S. technology from indirectly supporting China’s armed forces.
The developments underscore the difficult balance between commercial interests and national security in high-tech exports, particularly in sectors such as AI where U.S. firms lead globally. As Nvidia continues to engage with Chinese customers, the scrutiny from U.S. lawmakers signals a growing political and regulatory focus on the potential military implications of commercial AI collaborations.
