Nvidia is facing new scrutiny in Washington after a senior U.S. lawmaker claimed that its technical assistance helped Chinese startup DeepSeek refine artificial intelligence models that were later tied to Chinese military applications.
According to a letter reviewed by a prominent publication, Rep. John Moolener, the Republican chairman of the House Select Committee on China, claimed that NVIDIA engineers played a role in improving the efficiency of DeepSeek’s AI training system. The development has fueled concerns that U.S. technology could inadvertently accelerate China’s defense capabilities, despite strict export controls.
DeepSeek gained global attention last year when it unveiled an AI model that it says rivals leading U.S. systems while requiring far less computing power. The breakthrough surprised industry experts and policymakers alike, especially as the United States tightens restrictions on sales of advanced semiconductor hardware to China to maintain its technological edge.
In a letter to U.S. Secretary of Commerce Howard Lutnick, Moolener said NVIDIA’s internal records reveal close collaboration with DeepSeek’s technology team.
“NVIDIA records show that NVIDIA technology development personnel have helped DeepSeek significantly improve its training efficiency through ‘optimized co-design of algorithms, frameworks, and hardware,’ and internal reports state that ‘DeepSee “K-V3 boasts that full training requires only 2.788 million H800 GPU hours, which is less than what U.S. developers typically require for frontier-scale models,” Moolenaar wrote in the letter.
GPU time measures the time required for an AI chip to train a complex model. Frontier scale systems refer to top-of-the-line models built by companies like OpenAI, Anthropic, and Google.
The documents cited by the committee relate to NVIDIA’s activities during 2024. At the time, there were no clear public indications that DeepSeek’s activities might support the Chinese military. Mr. Moulenard acknowledged this background in his letter, writing:
“NVIDIA has properly treated DeepSeek as a legitimate commercial partner deserving of standard technical support,” Moolenaar wrote.
DeepSeek reportedly used Nvidia’s H800 chip, a processor designed specifically for the Chinese market, before it was added to the US export control list in 2023. U.S. officials have previously suggested that the company may now be contributing to China’s defense-related research.
In response to the allegations, Nvidia emphasized that China has sufficient domestic semiconductor manufacturing capacity and downplayed the possibility of military dependence on US technology.
“China has enough domestically produced chips for all military applications, with millions to spare. It is nonsense for the Chinese military to rely on American technology, just as it is nonsense for the US military to use Chinese technology,” Nvidia said in a statement.
The issue resurfaced recently when the Trump administration allowed limited sales of Nvidia’s more powerful H200 chip to China under strict conditions that prohibited sales to military organizations. The decision has drawn criticism from lawmakers who argue that enforcing such restrictions remains difficult.
“When even the most valuable companies in the world cannot exclude military use of their products when selling to businesses, strict licensing restrictions and enforcement are essential to prevent such warranties from becoming superficial formalities,” Moolenaar wrote.
He added: “Chip sales to ostensibly non-military end users in China would necessarily violate military end-use restrictions.”
The Commerce Department, the Chinese Embassy in Washington, and DeepSeek had no immediate comment, leaving the debate over AI technology transfer and national security far from settled.
