China's Z.ai reportedly powers up 1GW data center built entirely on domestic chips

Z.ai’s fully domestic-chip-powered data center marks China’s AI infrastructure entering a de facto 'de-Americanized' phase. Technically, this accelerates iteration in homegrown GPUs, high-speed interconnects, and liquid cooling—benefiting ecosystem players like Cambricon and Ascend—but software stack maturity remains a bottleneck. Compliance-wise, while firms sidestep U.S. export controls, they face higher operational costs and yield volatility. NVIDIA may temporarily retain market share via H20 variants but will likely restructure global product lines amid deepening tech bifurcation. Over the next 12–24 months, China’s AI compute market will split into dual tracks: domestic stacks dominate government and SOE workloads, while international chips serve foreign-invested and outbound-focused enterprises. This fragmentation inflates IT spending and could catalyze a CUDA-alternative software standard, reshaping the global AI foundation layer.