DeepSeek, the Chinese startup behind large language models that rival those from US firms such as OpenAI and Anthropic, is planning to design its own silicon, according to a Reuters report citing three people familiar with the matter. The move matters because it reflects how US chip export controls are pushing Chinese AI companies to seek greater self-sufficiency in the hardware that underpins their models, reducing their dependence on outside suppliers.

DeepSeek has reportedly been working on the effort for about a year, meeting potential hardware partners and hiring engineers, with a focus on data-centre chips for inference rather than training. The aim is said to be reducing reliance on both Nvidia — largely shut out of China by a US export ban — and Huawei, which controls roughly half of China's data-centre chip market. DeepSeek is not alone, with Alibaba and Baidu also moving into chips. The trend extends to the US, too: OpenAI and Broadcom recently unveiled an inference chip called Jalapeño, and Anthropic has been exploring custom designs, partly to cut reliance on Nvidia and gain tighter control over their technology stacks amid constrained data-centre capacity.