DeepSeek’s 75-million open source downloads are driving China’s AI hardware pivot, giving Huawei early access while sidelining Nvidia and AMD to cut reliance on US chips.
Chinese AI lab DeepSeek is turning open-source momentum into hardware leverage, with more than 75 million downloads of its models on Hugging Face helping Chinese AI releases surpass every other country on the platform.
The scale of adoption is fuelling a new wave of Chinese open-source systems competing directly with US labs and intensifying debate in Washington over exports of advanced AI processors to China.
Against that backdrop, DeepSeek has broken with industry norms ahead of its flagship V4 release. Instead of sharing pre-release access with long-time partners Nvidia and Advanced Micro Devices for optimisation, the lab granted early access to domestic suppliers, including Huawei Technologies, giving Chinese chipmakers several weeks to tune software for local hardware.
AI developers typically collaborate closely with US chipmakers to ensure peak performance. DeepSeek itself previously worked with Nvidia engineers, making the exclusion a notable departure.
Geopolitical tensions are adding to scrutiny. A senior Trump administration official told Reuters the latest model was trained on Nvidia’s Blackwell chips in mainland China, potentially breaching export controls. DeepSeek may remove technical indicators of US hardware and claim Huawei chips were used instead.
Market impact appears limited for now. As Ben Bajarin, CEO of Creative Strategies, said: “The impact to Nvidia and AMD for general data accelerators is minimal – most enterprises are not running DeepSeek, which serves as a benchmarking model more than anything else.”
He added tools are cutting optimisation time “from months to weeks,” and the move may aim “to try to keep U.S. hardware and models disadvantaged” in China.
