Chinese artificial intelligence lab DeepSeek has departed from industry convention by declining to share early access to its forthcoming flagship model, V4, with major US chipmakers Nvidia and Advanced Micro Devices. Instead, DeepSeek provided pre-release access to domestic suppliers, including Huawei Technologies, granting them a head start of several weeks to optimize performance on their processors. Artificial intelligence developers typically work closely with Nvidia and AMD before major launches so models run efficiently on widely used hardware, and DeepSeek had previously collaborated with Nvidia’s technical staff.
Nvidia and AMD did not comment on the decision, and DeepSeek and Huawei did not respond to inquiries. Analysts said the immediate commercial impact on Nvidia and AMD is likely limited because most enterprises are not running DeepSeek, which is described as serving more as a benchmarking model. Ben Bajarin, CEO of research firm Creative Strategies, said the shift fits a broader Chinese strategy “to try to keep U.S. hardware and models disadvantaged” in China, and noted that new artificial intelligence coding tools are shortening optimization work “from months to weeks.” The timing coincides with mounting US concern over how advanced American artificial intelligence chips are used in China.
A senior Trump administration official said DeepSeek’s latest artificial intelligence model was trained on Nvidia’s most advanced chip, Blackwell, using a cluster in mainland China, in a move that appears to violate U.S. export controls. According to the official, DeepSeek may attempt to remove technical indicators exposing its use of American chips and plans to publicly claim that Huawei hardware powered training. DeepSeek’s models have been downloaded more than 75 million times on open-source platform Hugging Face since its emergence in January 2025, and downloads of Chinese models released in the past year on the platform have surpassed those from any other country. The rapid ascent of Chinese open-source artificial intelligence has intensified debate in Washington over exporting advanced US artificial intelligence chips to China. U.S. authorities last year allowed Nvidia’s H20 and AMD’s MI308 chips – designed for artificial intelligence inference – to resume shipments to China, but licenses for more advanced processors remain restricted. The H20 and MI308 chips are aimed at inference, the process of running trained artificial intelligence models, and demand for the MI308 was significant, with AMD saying it generated 390 million in sales of the chip in its most recent quarter. DeepSeek is one of several Chinese artificial intelligence firms expected to roll out new models this month.
