DeepSeek is secretly developing its own AI inference chips

Odaily reported that sources familiar with the matter have revealed that Chinese AI large model company DeepSeek is secretly developing AI inference chips. The relevant project was initiated approximately one year ago and is primarily intended for model inference rather than training. According to reports, DeepSeek has been continuously recruiting chip design engineers in recent months to advance this project.

Currently, DeepSeek's model training and inference mainly rely on chips from Nvidia and Huawei. If the self-developed inference chips are successfully launched, the company could reduce its dependence on external suppliers and improve control over hardware costs. However, the project is still in its early stages and also faces restrictions on chip manufacturing and memory access imposed by U.S. export controls. (Reuters)