DeepSeek works on its own inference chip

DeepSeek is developing its own AI chip, according to three people familiar with the matter, in a move that could reduce its dependence on Nvidia and Huawei hardware. The chip is being designed for inference, the stage when a trained model generates responses for users, rather than for training new models.

The push would mark a strategic shift for a company better known for AI model breakthroughs than commercialising its technology. A US ban on Nvidia exports to China has helped Huawei gain around half of the US$50 billion (S$65 billion) domestic AI chip market, but Huawei is already facing pressure from Alibaba and Baidu, which are developing their own chips.

DeepSeek’s project remains at an early stage, with outreach to chip-design, foundry and memory companies. The company has also privately increased hiring of chip-design engineers in recent months. DeepSeek founder Liang Wenfeng said in a rare 2024 interview that chip export controls were a challenge for the company.

Manufacturing could be a major obstacle because US restrictions limit Chinese access to advanced overseas foundries and high-bandwidth memory. DeepSeek’s chip effort comes as the company was slated to raise US$7 billion in a maiden funding round valuing it at between US$52 billion and US$59 billion.