Representative John Moolenaar, a Republican who chairs the House Select Committee on China, wrote to the U.S. Commerce Secretary claiming internal Nvidia documents show company engineers helped Beijing-based DeepSeek optimise its AI systems.

According to the letter, the work focused on squeezing efficiency out of DeepSeek’s training processes, using fewer GPU hours, or computing cycles on specialised chips, to achieve performance rivalrous with leading U.S. models from firms such as OpenAI or Google.

The letter cites internal reporting that DeepSeek’s V3 model required about 2.788 million hours on Nvidia’s H800 GPUs for training, substantially below typical levels for so-called frontier AI models. Nvidia’s H800 chips were sold in China before tighter U.S. export controls came into effect in 2023.

DeepSeek became a focal point last year after it stunned markets by producing high-performing AI models with surprisingly modest compute, fuelling concern in Washington that China could close the gap with U.S. industry despite export curbs. U.S. officials have previously said DeepSeek’s technology was aiding China’s military, though the company has not publicly confirmed this.

In his letter, Moolenaar notes Nvidia treated DeepSeek as a standard commercial customer at the time and was not publicly aware of any military use. But the claim has reignited debate over how effective U.S. export controls and licensing rules are at preventing advanced AI expertise from indirectly bolstering foreign military capabilities.

Nvidia has argued it sold chips designed for the Chinese market and that assertions about reliance by China’s military misunderstand China’s own domestic chip base. The U.S. Commerce Department and DeepSeek did not immediately comment.

The issue comes as U.S. policy on AI chip exports remains in flux. Earlier this month, the U.S. approved limited sales of more advanced Nvidia H200 chips to China with conditions aimed at barring their use by military-linked entities, a move that has drawn criticism from some lawmakers afraid technology could still be diverted.

RELATED STORIES

Microsoft report warns of DeepSeek’s meteoric rise and a global AI divide

China’s DeepSeek breaks silence with rare public warning on AI disruption