Z.ai (formerly Zhipu AI) has completed a massive 1-gigawatt compute facility powered entirely by Chinese-made accelerators, with no Nvidia chips, and has already partially commissioned it for training next-generation GLM models, according to Bloomberg. The company also operates several additional clusters of over 10,000 chips each. While this doesn’t prove domestic chips match Nvidia on performance or ecosystem, it demonstrates a viable alternative compute platform is forming — as Beijing considers roughly $295 billion in investment for a national data-centre network requiring at least 80% domestic technology.

Bloomberg — Z.ai Completes Giant Data Center With Chinese Chips to Train AI