Nvidia Aided DeepSeek’s AI Model Despite Export Controls, US Lawmaker Alleges

Nvidia Corp. provided technical support enabling Chinese startup DeepSeek to achieve near-state-of-the-art AI performance using restricted H800 chips, undermining U.S. export controls, according to Rep. John Moolenaar (R-MI) in a Jan. 29, 2026 letter to Commerce Secretary Howard Lutnick. Internal Nvidia records show DeepSeek-V3 required just 2.8 million H800 GPU hours for training, leveraging “co-designed” algorithms and frameworks. Nvidia also proposed packaging DeepSeek as an enterprise product for its hardware. The H800, a downgraded H100 variant, was permitted for sale to China until October 2023. Communications between the firms spanned June 2024 to May 2025. Moolenaar urged strict enforcement of H200 export rules following Trump’s December 2025 policy easing. Nvidia countered that restricting sales aids foreign competitors and stressed China has ample domestic chips for military use. DeepSeek’s R1 model triggered a 3% Nasdaq 100 drop in 2025 amid cost-efficiency concerns.