DeepSeek is now reportedly designing its own chip: a bid to cut Nvidia reliance

DeepSeek, the Chinese AI startup, is reportedly working on its own inference chip, the part that spits out answers when you ask a model something, Reuters reports. If that holds, it’s a real change in direction: less dependence on Nvidia, and to a lesser degree, on Huawei hardware too.

Reuters says DeepSeek is aiming at inference, not training. That tracks. Serving chatbot responses at scale burns through money fast, and a custom chip could give the company tighter control over cost, performance, and how those systems are deployed across the whole stack.

That push toward self-reliance looks even more urgent now that Washington has tightened export controls. Reuters says DeepSeek had been using Nvidia’s China-specific H800 chips before those also got caught by additional U.S. restrictions, which makes access to top-end compute a lot less predictable for Chinese AI companies.

If you pay attention to AI infrastructure, this is one to keep an eye on. Still, it’s early.

Reuters says DeepSeek is in talks with chip designers, foundries, and memory suppliers, while also quietly hiring engineers. Even so, a processor that can really compete would take years, a lot of money, advanced manufacturing capacity, and high-bandwidth memory, all of which are still vulnerable to bottlenecks and export limits.

Per Reuters, DeepSeek would be joining a broader push that’s already underway, one that includes major AI companies and Chinese players like Alibaba, Baidu, and Huawei. For now, this looks like a long-term bet on survival, independence, and control, not a product launch that’s right around the corner.