DeepSeek moves to develop its own AI inference chip to cut reliance on Nvidia and Huawei

Chinese AI startup DeepSeek is engineering a custom chip aimed at the inference stage of AI models – the process that generates responses from already‑trained models – in order to lessen dependence on Nvidia GPUs and Huawei’s Ascend processors. Reuters reports that the year‑long project is still in its early phase, with the company quietly hiring chip‑design engineers and engaging potential design‑foundry‑memory partners. The initiative mirrors a broader trend among AI firms, including OpenAI and Anthropic, to build proprietary silicon for cost, performance and supply‑chain control.

The move comes as the United States maintains export restrictions on advanced Nvidia chips to China, prompting Beijing to consider a limited easing of its own ban on Nvidia’s H200 GPUs for firms such as DeepSeek, Alibaba and ByteDance. Analysts note that news of DeepSeek’s chip plans caused Nvidia shares to slip about 1.6% in pre‑market trading. If successful, DeepSeek’s hardware could reshape China’s AI ecosystem, reduce reliance on external suppliers, and intensify competition for Nvidia and Huawei in the domestic market.