DeepSeek Is Designing Its Own AI Chip — and Taking Outside Money for the First Time

DeepSeek, the Chinese AI lab that shook the industry with its inexpensive open models, is designing its own AI chip, Reuters reports, citing three people familiar with the matter. The chip is built for inference — the phase where a trained model actually answers your questions — rather than for training new models from scratch.

The project is still early. DeepSeek is reportedly in talks with chip design, manufacturing, and memory companies, and has been quietly hiring chip engineers for months without posting public job listings. The goal, per the report: reduce its reliance on chips from Nvidia and from China’s own Huawei.

At the same time, DeepSeek is raising outside capital for the first time in its history — about $7 billion at a valuation between $52 and $59 billion. For a company that famously ran lean and self-funded, that’s a notable shift, and a chip program is exactly the kind of project that burns money at that scale.

What’s behind it? Two pressures point in the same direction. First, US export controls cut Chinese companies off from the most advanced chips and memory, so anything DeepSeek can do to control its own hardware supply reduces a strategic weakness. Second, this mirrors a broader industry pattern: OpenAI and Anthropic are also working on custom chips — OpenAI recently unveiled an inference chip with Broadcom. When your business is running models for millions of users, inference is where the ongoing costs live, and custom silicon tuned to your own models can cut that bill substantially. Nvidia’s chips are brilliant general-purpose tools, but you pay for that generality.

Worth keeping expectations grounded: designing a competitive AI chip typically takes years, and the report stresses the project is in its early stages. Plenty of chip ambitions have quietly died between announcement and silicon.

What this means for you: Nothing changes today, but the trend is worth understanding. Cheap inference is why DeepSeek’s models cost a fraction of what US rivals charge — and cheaper hardware would push those prices down further. If you use affordable AI models through apps or APIs, the chip race among AI labs is one of the forces working in your favor. It’s also a signal about where the industry thinks the real bottleneck is: not smarter models, but the cost of running them at scale.

Sources

• Reuters: China’s DeepSeek developing its own AI chip, sources say
• The Decoder: Deepseek is designing its own AI chip