Reuters Exclusive Report: DeepSeek to Develop Its Own Inference Chip
According to a Reuters exclusive report from July 7, 2026, the Chinese AI lab DeepSeek is working on its own chip for inference workloads – that is, for the computation phase in which a pre-trained model generates answers, not for training itself. The news agency cites three unnamed individuals with knowledge of the matter who wished to remain anonymous because the information is confidential. DeepSeek did not respond to a request for comment.
A Year in Secrecy, Now Increased Recruitment Efforts
The project has reportedly been underway for about a year, since around July 2025. In recent months, DeepSeek has been actively recruiting experienced chip engineers through private channels – for chip architecture, verification, and the associated software layer. According to TechNode, the team thus covers the entire development process, from chip design to software integration.
Reuters also reports that DeepSeek has already initiated discussions with chip design, foundry, and memory companies. Which companies are specifically involved – such as a foundry partner – is not mentioned in any of the analyzed reports.
Why an Own Chip? Costs and Dependence on Nvidia and Huawei
Computing power is one of the largest cost blocks for AI companies – according to industry estimates, sometimes over half of total operating costs. An own chip tailored to its models could reduce these costs.
Additionally, there is the geopolitical situation: DeepSeek initially relied on Nvidia’s throttled H800 chips for the Chinese market. According to SiliconANGLE, these chips were banned due to tightened US export controls, prompting DeepSeek to increasingly turn to Huawei’s Ascend graphics processors. The Reuters sources now report growing concern about becoming too dependent on a single supplier – an own chip would be a third, independent option. DeepSeek founder Liang Wenfeng had already admitted in an interview in 2024 that chip export controls pose a challenge for the company, suggesting that the topic has been discussed internally for some time.
Not an Isolated Case: OpenAI, Alibaba, and Baidu Also Building Their Own Chips
With this step, DeepSeek joins a broader industry trend. OpenAI introduced its first own chip in June 2026 together with Broadcom, the inference processor “Jalapeño.” The goal there is also to reduce dependence on Nvidia and gain more control over the entire technology stack – a strategy reminiscent of Apple’s vertical integration of hardware and software.
According to SiliconANGLE, Alibaba and Baidu are also working on their own AI processors. Semafor additionally reports that Huawei claims about half of the Chinese AI chip market this year – another reason why Chinese AI labs are seeking alternatives to a single dominant supplier. Bloomberg confirms the Reuters report and places it in the same context: reducing dependence on Nvidia and Huawei in the training and operation of DeepSeek’s models.
Financing: Conflicting Figures, No Independent Confirmation
Different figures are circulating regarding the financial backing of the chip project. TechNode reports that DeepSeek is currently seeking $7 billion in external capital at a valuation of $52 to $59 billion. Other reports refer to a completed funding round from Chinese investors amounting to $7.4 billion. Both figures come from reports with anonymous sources and are not independently verified; it remains unclear whether they refer to the same or different funding rounds.
Timeline: Still Far from Mass Production
Even if the project proceeds as planned, industry observers estimate that it will take significantly longer than a year for a chip design to become a market-ready product. As of the current knowledge, DeepSeek is thus in an early development phase – without a confirmed foundry partner and without a publicly communicated timeline.
Context
The report confirms a development that has been evident since the first US export restrictions on AI chips: Major AI providers – whether American or Chinese – increasingly want to gain control over the hardware layer itself, rather than relying on a single external supplier. For DeepSeek, which is among the most well-known Chinese AI labs with its models V3 and R1, an own inference chip would be another building block on the path to greater technological independence – provided that the early-running project actually reaches mass production.
