Overview

• DeepSeek’s tests on Huawei’s Ascend AI chips failed to stabilize its R2 model due to hardware and interconnect performance limitations.
• The company plans to retrain R2 on Nvidia’s H20 accelerators, highlighting its reliance on the CUDA ecosystem and proven software support.
• The pivot challenges Beijing’s drive for domestic chip usage after U.S. export controls restricted China’s access to foreign processors.
• The R2 rollout, initially expected in May, now faces an open-ended timeline as technical setbacks delay rollout plans.
• Industry analysts warn that persistent domestic chip gaps will prolong China’s dependence on Nvidia and could reshape regional hardware demand.