DeepSeek is reportedly designing its own AI inference chip to be manufactured by SMIC, China's largest foundry, according to an anonymously sourced Reuters report. The move would reduce the lab's dependence on Nvidia hardware restricted by US export controls. The chip targets inference workloads, where roughly 70% of AI compute demand now lies and where Chinese silicon is most competitive. Significant obstacles remain: SMIC is limited to a 7nm process several generations behind TSMC, yields are reportedly low, and DeepSeek's own R2 model was delayed due to failures on Huawei hardware. With a reported $45bn state-backed funding round, money is no longer the bottleneck, but turning a chip design into a shipping product is a multi-year challenge that DeepSeek has not confirmed pursuing.
