Chinese models reportedly used banned Nvidia chips
By the AFREEQUE newsroom, based on TechCrunch
Open-access Chinese artificial intelligence models were reportedly trained with Nvidia chips whose export to China is restricted, according to an analysis by Will DePue. The apparent use of Blackwell processors by several Chinese laboratories highlights the difficulties of enforcing US export controls, reports TechCrunch.
Researchers say several major open-source artificial intelligence models developed by Chinese companies show traces of training on Nvidia Blackwell processors whose export to China is banned by the United States. Will DePue, a former OpenAI researcher who heads the non-profit research group Unemployment Labs, analysed the parameters of models from DeepSeek, Moonshot AI, MiniMax and Ant Group, reports TechCrunch.
According to this analysis, Moonshot AI’s Kimi K3 versions and DeepSeek’s V4 models notably have characteristics consistent with training on Blackwell chips. White House technology adviser Michael Kratsios had also said earlier this year, citing US intelligence sources, that Moonshot AI had used Nvidia chips subject to restrictions, reports TechCrunch.
The organisation Epoch AI estimates that around two million advanced chips entered China in 2024 and 2025 through schemes designed to conceal their final destination. It also estimates that 275,000 Nvidia AI accelerators were imported into Malaysia, where Chinese companies can access computing capacity without directly importing the chips into China, according to TechCrunch.
These elements do not, however, establish that the models concerned were trained entirely on Nvidia chips or that US export rules were necessarily violated, reports TechCrunch.
