Chinese AI startup DeepSeek is developing its own AI chip, Reuters reported on July 7, citing three people familiar with the matter. The chip is designed for inference — the stage where a trained model generates responses for users — rather than for training new models. If successful, the move would reduce the company’s dependence on both Nvidia and Huawei, and mark a major strategic shift for the startup widely regarded as China’s AI champion. AINsf breaks down what is known.
An early-stage effort, hired in secret
According to Reuters’ sources, the project began about a year ago and remains at an early stage: the Hangzhou-based company has been holding discussions with chip-design firms, foundries, and memory suppliers. DeepSeek has also stepped up hiring of chip-design engineers in recent months — but quietly, without posting the roles on public job platforms. The company did not respond to requests for comment.
The timing is notable for another reason: the chip push coincides with DeepSeek’s first-ever embrace of outside capital. Reuters reported in June that the company was set to raise $7 billion in a maiden funding round at a valuation between $52 billion and $59 billion, reversing its years-long strategy of rejecting external investment.
Why inference, and why now
Inference is the fastest-growing segment of AI computing demand. As AI applications spread, more of the industry’s workload is shifting from training models to running them — a job that specialized chips can do more cheaply and with less power than general-purpose GPUs. DeepSeek would be joining a broader industry trend: OpenAI unveiled Jalapeno, its first custom inference chip built with Broadcom, in June, and Anthropic has been weighing its own silicon, Reuters reported in April.
For DeepSeek, though, the effort carries an extra geopolitical dimension. US export controls bar Chinese companies from buying Nvidia’s most advanced chips, and Beijing has been pressing its tech champions to build domestic alternatives. The foundation model behind R1 — the reasoning model whose low-cost performance triggered a rout in US tech stocks in January 2025 — was trained on Nvidia’s H800, a China-market chip Washington banned in late 2023. Since then DeepSeek has leaned increasingly on Huawei, releasing its V4 model adapted for Huawei’s Ascend processors in April. A homegrown DeepSeek chip would therefore squeeze not only Nvidia but also Huawei, which has captured roughly half of China’s $50 billion domestic AI chip market since the US export ban.
The obstacles are real
Designing a competitive AI chip typically takes years and enormous capital, and manufacturing is the harder problem: US restrictions bar Chinese designers from the world’s most advanced foreign foundries, and separate curbs have cut China’s access to high-bandwidth memory, a component critical for inference chips. Analyst Richard Windsor of Radio Free Mobile was blunt about the market implications, telling Reuters: “Nvidia is at zero in China and staying there” — and arguing DeepSeek has little chance of selling silicon outside China without access to leading-edge manufacturing.
Even so, the direction of travel is unmistakable: the world’s top AI labs — American and Chinese alike — no longer want to rent the hardware their futures run on.