DeepSeek denied early access to Nvidia and AMD while giving Huawei a head start, signalling how open source Chinese AI is reducing reliance on US chips and weakening export leverage.
Chinese AI lab DeepSeek has withheld its upcoming V4 model from Nvidia and Advanced Micro Devices, breaking the long-standing industry practice of sharing pre-release builds with leading chipmakers for performance optimisation.
Instead, the company granted early access to domestic suppliers, including Huawei Technologies, giving Chinese processors a weeks-long head start to tune the software.
The move suggests a deliberate shift to strengthen China’s local silicon ecosystem while disadvantaging US hardware vendors, at a time when Washington is tightening export controls on advanced AI chips. It also raises compliance questions after a US official said DeepSeek’s latest model may have been trained on Nvidia’s Blackwell chips inside China.
DeepSeek’s growing influence stems from open distribution. Its models have logged more than 75 million downloads on Hugging Face, fuelling a wave of Chinese open-source releases that now outpace other countries on the platform.
The portability of open models, combined with faster optimisation tools, reduces dependence on vendor-specific support and weakens traditional chip leverage.
“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,” said Ben Bajarin, CEO of Creative Strategies. He added that tools are cutting optimisation times “from months to weeks,” and the strategy aims “to try to keep US hardware and models disadvantaged” in China.
Nvidia, AMD, DeepSeek and Huawei declined to comment. The reason for the access decision remains unclear.
