Chinese AI start-up DeepSeek is developing its own processor for the model industry, sources told Reuters. Work began about a year ago, the project is in its early stages: the company is discussing chip design, contract production and memory with external partners, and is hiring closed-loop semiconductor design engineers.

The goal is to reduce dependence on Nvidia and Huawei. Previously, DeepSeek used the Nvidia H800 (a special version for China) to train the base R1 model, then switched to Huawei’s Ascend chips to adapt the V4 model. Development is taking place against the backdrop of US export restrictions and growing demand for industry solutions. Founder of DeepSeek Liang Wenfeng in 2024 called chip export controls a serious problem.

The race for its own AI infrastructure is led by OpenAI, Broadcom, Anthropic, Alibaba and Baidu. Reuters notes that DeepSeek’s success is not guaranteed due to high competition, significant investment and limited access to advanced overseas factories. At the same time, the company is forming a team to develop Code Harness - an autonomous programming tool that competes with Claude Code and Codex.