China's DeepSeek is developing its own AI chip in a push to reduce its reliance on both Nvidia and Huawei, according to an exclusive Reuters report citing people familiar with the effort. The Hangzhou-based company has quietly increased its hiring of chip-design engineers over roughly the past year, largely through private channels rather than public job postings, and is holding early-stage discussions with external chip-design, foundry and memory partners.
The chip is being designed for inference -- the stage of AI computing where a trained model generates responses for users -- rather than for training new frontier models. Inference is the fastest-growing segment of AI compute demand, and specialized silicon can run meaningfully cheaper and more power-efficiently than general-purpose GPUs at that stage, which is why DeepSeek, OpenAI, Google and Amazon have all pursued custom inference silicon rather than relying solely on merchant GPU vendors.
“The chip is being designed for inference -- the stage of AI computing where a trained model generates responses for users -- rather than for training new frontier models.”
Success would position DeepSeek as a new domestic rival to Huawei inside China's roughly $50 billion AI chip market, while reducing its dependence on Nvidia GPUs that remain subject to US export controls limiting China's access to the most advanced chips. For a company whose entire public identity is built on doing more with cheaper compute, in-house inference silicon is a logical extension of that positioning.
But the manufacturing path is genuinely difficult. US export rules bar Chinese chip designers from using the most advanced overseas foundries, and separate curbs have restricted China's access to high-bandwidth memory, a component that's central to competitive inference-chip performance. DeepSeek's effort remains early-stage, and Reuters' sources caution the company is still reaching out to potential partners rather than having locked in a manufacturing path.
The bear case: chip design is a multi-year, capital-intensive undertaking even before manufacturing constraints are factored in, and DeepSeek is simultaneously trying to close a new funding round at a ~$71 billion valuation and prepare a Shanghai IPO -- a lot of major initiatives running in parallel for one company. What to watch next: whether DeepSeek names a foundry partner, and whether Beijing's broader semiconductor self-sufficiency push (evident in CXMT's imminent Shanghai IPO) extends state support to DeepSeek's chip ambitions specifically.