WASHINGTON, Jan 29, 2026, 10:25 (EST)
- House China committee chair says Nvidia support helped DeepSeek sharpen AI later used by China’s military
- Nvidia rejects the idea Beijing’s military depends on U.S. tech; Commerce Department and DeepSeek did not comment
- CEO Jensen Huang says China is still finalising an import licence for Nvidia’s H200 chip
A U.S. House panel leader accused Nvidia of helping Chinese AI company DeepSeek sharpen its models, which he said were later used by China’s military, and urged tighter curbs on sensitive chip exports. Representative John Moolenaar, who chairs the House Select Committee on China, raised the issue in a letter to Commerce Secretary Howard Lutnick. (Reuters)
The letter lands as President Donald Trump’s administration has approved exports of Nvidia’s H200 chip to China with restrictions meant to keep it away from military-linked users, a policy already drawing criticism from China hawks.
DeepSeek is a sore point in Washington because its models rattled markets last year by matching leading U.S. offerings while using less computing power. That cuts against the logic of export controls — U.S. limits on what tech can be shipped abroad for national security reasons — if China can do more with fewer chips.
In his letter, Moolenaar said Nvidia staff helped DeepSeek make an “optimized co-design of algorithms, frameworks, and hardware” that improved training efficiency. He cited internal reporting that DeepSeek-V3 trained using 2.788 million H800 GPU hours — how long graphics processing units, the chips used to train AI, have to run — less than U.S. developers typically need for top-end systems.
He wrote the work took place in 2024, before there was public evidence DeepSeek’s technology was being used by China’s military. Nvidia’s H800 was designed for the China market and sold there before it fell under U.S. export controls in 2023.
Nvidia said it would be unreasonable for China’s military to rely on American technology, arguing China had enough domestic chips for military uses. China’s embassy in Washington said Beijing opposed stretching the concept of national security to politicise trade, and the Commerce Department and DeepSeek did not respond to requests for comment.
Nvidia CEO Jensen Huang, speaking in Taipei after a trip to China, said “the actual license for H200 is being finalised” and he hoped for a favourable decision from Beijing. The H200 is more powerful than the H800 and has become a flashpoint in U.S.-China tech ties. (Reuters)
China has already given conditional approval to ByteDance, Alibaba and Tencent to buy more than 400,000 H200 chips, four people familiar with the matter told Reuters, though some terms were tight enough that buyers have held off converting them into orders. The same people said Chinese officials have discussed requiring companies to pair imports with domestic chips, as local supplier Huawei pushes alternatives that still lag the H200. (Reuters)
Olivia Shen, director of strategic technologies at the United States Studies Centre at the University of Sydney, said export controls look neat on paper until the hardware leaves the country: “it’s incredibly difficult to control how the chips will actually be used downstream.” She said the timing matters because the first H200 shipments have been cleared for export, and Nvidia is investing in location-verification tools to limit smuggling risks. (Taiwanplus)
The next fight is over enforcement, not engineering. Lawmakers could press Commerce to tighten licences or broaden the definition of military end use, while China could slow approvals to protect its domestic chip industry — either way, Nvidia risks getting squeezed between two governments.
Moolenaar warned that sales to nominally civilian customers in China would still “inevitably” breach military end-use restrictions, calling for tougher licensing and follow-up checks.
