DeepSeek is working on its own inference chip as China’s AI darling tries to loosen Nvidia and Huawei’s grip.
According to Reuters, the move could cut its reliance on Nvidia and Huawei chips, which it has used to train and run its globally popular models.
The chip is designed for inference, the stage of AI computing where a trained model generates responses for users. It is not meant for training new models, the sources said.
If it works, DeepSeek’s move into semiconductors would mark a serious strategic shift for a company hailed in China as its AI champion.
It could create another headache for Chinese tech giant Huawei, which already has rivals chewing at its domestic AI chip business. Shares of US-based Nvidia slipped about 1.6 per cent in premarket trading.
Radio Free Mobile analyst Richard Windsor told Reuters: “Nvidia is at zero in China and staying there. DeepSeek has almost no chance of selling silicon outside of China unless it gets access to leading-edge manufacturing.”
DeepSeek became famous more than a year ago after releasing two efficient AI models that went viral worldwide. The launch surprised plenty of folk in Silicon Valley and Washington.
The company has long focused on AI model breakthroughs rather than squeezing cash from its technology.
Huawei’s kit still trails Nvidia’s most advanced chips by a wide margin. However, a US export ban on Nvidia’s top chips to China helped Huawei grab about half of the $50 billion domestic AI chip market.
Huawei has supplied DeepSeek and several other leading industry players. That cosy position is weakening as Alibaba and Baidu develop their own AI chips and win market share.
DeepSeek’s attempt to join the silicon race is still in its early stages. The company has contacted external partners and held talks with chip design, foundry, and memory companies, the three sources said.
The Hangzhou-based company has increased hiring of chip-design engineers in recent months. Recruitment has been handled privately, without job postings on public hiring platforms.
DeepSeek would be following other global AI developers seeking more control over the hardware behind their models and less dependence on Nvidia.
DeepSeek founder Liang Wenfeng said in a rare 2024 interview with Chinese media that chip export controls were a challenge for the company. DeepSeek has used both Nvidia and Huawei chips. The company has said the foundation model behind R1 was trained on Nvidia’s H800.
R1 was the reasoning model whose cheap performance helped trigger a rout in US tech stocks in January 2025. The H800 was designed for the Chinese market, but Washington banned it in late 2023.
DeepSeek has since leaned harder on Huawei. In April it released its V4 model adapted for Huawei’s Ascend chips. Huawei said its processors were used in part of the training of V4-Flash, a lighter version of the model.
