China aims to significantly increase its domestic AI chip production, potentially tripling output, as US restrictions limit access to Nvidia's high-end processors. Fabrication plants supporting Huawei are reportedly ramping up production to meet the growing demand. This push is partly driven by the need to support local AI models, such as those developed by DeepSeek.
DeepSeek's new V3.1 model is designed to run on next-generation domestic processors, utilising the UE8M0 FP8 data format to optimise memory efficiency and computational performance. This move signifies a shift towards integrated development of hardware and software within China's AI sector. While Nvidia chips remain dominant, China's substantial investment in its semiconductor industry could accelerate domestic innovation and reduce reliance on foreign technology. Huawei's Cloud Matrix architecture, combined with its Ascend chip series, has shown the ability to run DeepSeek's models efficiently.
Despite progress, challenges remain, including issues with the stability, connectivity, and software quality of Chinese chips compared to Nvidia. Nevertheless, the focus on domestic AI chip production reflects China's determination to achieve self-sufficiency in critical technologies and compete in the global AI landscape.
