Chinese labs test local chips for AI training

Chinese AI labs are increasingly testing whether domestic chips can support parts of model development often associated with Nvidia hardware. The shift is focused on moving earlier training phases onto locally made silicon, rather than demonstrating a complete replacement for Nvidia across the full model life cycle.

5 AI models have been trained on local chips, highlighting a broader effort within China’s AI sector to evaluate homegrown hardware for advanced development work. The experiments point to growing interest in reducing reliance on foreign processors while measuring how far domestic silicon can go in supporting model training.