Chinese AI startup DeepSeek is reportedly designing its own AI chip focused on inference—the stage where trained AI models generate responses for users. According to reports, the chip is being developed with China's leading contract chipmaker SMIC, marking a strategic move toward greater control over its AI infrastructure amid tightening U.S. export restrictions. The project is still in its early stages, and no working chip has been publicly demonstrated yet.
The custom chip is expected to reduce DeepSeek's dependence on both Nvidia GPUs and Huawei's Ascend processors. While DeepSeek has relied on Nvidia hardware for training and Huawei chips for some inference workloads, building an in-house accelerator could lower operating costs, improve performance for its own AI models, and reduce exposure to supply chain disruptions caused by export controls.
Developing a competitive AI chip, however, remains a major challenge. China's semiconductor industry continues to face restrictions on access to advanced manufacturing technology and high-bandwidth memory, making it difficult to match the capabilities of leading AI chips produced by Nvidia. As a result, experts caution that designing a chip is only the first step, and successfully bringing it into large-scale production will require overcoming significant technical and manufacturing hurdles.
If successful, DeepSeek would join a growing list of AI companies—including OpenAI, Anthropic, and major cloud providers—seeking to develop custom AI hardware tailored to their own workloads. The initiative reflects a broader industry trend toward vertically integrated AI infrastructure, where companies build both the software and the silicon that power their models, while also advancing China's goal of greater technological self-reliance.
