China's Ascend and Cambricon chips drive shift to domestic AI compute for open-weight models

AI Impact Summary

The piece documents a strategic shift from NVIDIA-dominated compute toward domestic Chinese chips (Huawei Ascend, Cambricon, Baidu Kunlun) powering both inference and training for open-weight models such as Qwen, GLM and Kimi. For engineers, this implies multi-architecture workloads, platform-specific optimizations, and tighter coupling to chip vendors as software ecosystems diverge from CUDA-centric tooling. Businesses should plan for cost, supply risk, and vendor lock-in changes, with migration paths increasingly centered on Ascend/Cambricon-enabled pipelines and open-weight deployments.