DeepSeek is working on AI chips as export curbs limit Nvidia access
The Chinese AI startup is pursuing inference chips to reduce dependence on outside suppliers, Reuters reported.
By Maya Lindqvist · Senior Technology Correspondent
3 min read
DeepSeek is preparing a move into AI chip design as U.S. export restrictions continue to limit Chinese companies’ access to Nvidia hardware, Reuters reported. The effort matters because chips for running AI models have become a strategic bottleneck for companies building large language models.
Reuters, citing three people familiar with the matter, reported that DeepSeek has been working toward a silicon push for about a year. The company has held discussions with possible partners in hardware and chip development and has been recruiting engineers for the project, according to Reuters.
The work is focused on data center chips used for inference, Reuters reported. Inference is the stage in which a trained AI model processes user requests, a central workload for chatbots and other AI services.
DeepSeek’s likely aim is to reduce its dependence on Huawei and Nvidia, according to Ars Technica’s account of the Reuters report. Nvidia supplies much of the AI hardware used by companies in North America and Europe, but U.S. export controls have limited its ability to sell its most advanced chips in China.
China’s AI chip market is shifting
Huawei holds roughly half of China’s data center chip market, according to Ars Technica. DeepSeek would not be alone in trying to build or secure more domestic AI chip capacity: Alibaba and Baidu have also been pursuing chip-related efforts, Ars Technica reported.
The pressure on DeepSeek comes as Chinese AI developers seek more control over computing resources. Large language models require substantial data center capacity, and the hardware used to run them has become a key constraint for companies trying to expand AI products.
DeepSeek is known for developing large language models that compete with systems from U.S. companies such as OpenAI and Anthropic, according to Ars Technica. Its reported chip plans would extend that competition beyond software and model development into the hardware needed to operate AI services at scale.
U.S. AI firms are also looking beyond Nvidia
DeepSeek’s reported plan fits a broader industry pattern. Ars Technica reported that U.S.-based AI companies have also been pursuing custom chip designs, in part to reduce reliance on Nvidia.
OpenAI and Broadcom recently announced Jalapeño, OpenAI’s first chip designed for large-scale inference, according to Ars Technica. Anthropic has also explored custom chip design, though Ars Technica reported that no public milestones have been disclosed.
For OpenAI, Ars Technica reported, custom silicon is partly about reducing dependence on Nvidia and partly about gaining tighter control over the technology stack behind its products. Access to data centers and computing power is expected to remain constrained as AI companies compete to expand models and services, Ars Technica reported.
Reuters did not report a launch date for DeepSeek’s chip project in the details described by Ars Technica. The reported effort still signals that major AI developers in both China and the United States are treating chip supply as a core part of their competitive strategy.
This story draws on original reporting from Ars Technica.