Z.AI, formerly Zhipu, has built and begun operating a massive data center that runs entirely on domestically produced Chinese chips, with no Nvidia hardware involved, signaling a move toward self-reliant AI compute.

The facility is designed at about one gigawatt of power, housing multiple clusters with more than 10,000 chips each, and capable of supporting Z.AI’s most advanced GLM systems.

Industry observers highlight that this showcases hardware-software co-optimization in AI work, even as chip-level performance remains a consideration.

Analysts see broader implications for global AI compute dynamics and the competitive standing of Chinese models like Z.AI’s GLM and Moonshot open models.

Caveats include that Z.AI has not publicly confirmed chip specifics, and the efficiency of training on a mainly non-Nvidia cluster versus Nvidia systems remains unproven.

Bloomberg notes that exact chip identities were undisclosed, with local competitors such as Huawei, Cambricon, and Alibaba in the mix for Chinese AI accelerators.

The new facility is viewed as a test of whether China’s domestic chip industry can sustain frontier-model development amid global competition.

Z.AI recently listed in Hong Kong, signaling strong revenue expectations and investor confidence as it targets roughly $1 billion in annual recurring revenue.

The project represents a tangible move toward a Chinese-made, gigawatt-scale AI infrastructure that could influence export controls and the global hardware race.

Z.AI operates computing clusters with more than 10,000 chips each, indicating ongoing expansion in a competitive domestic AI development scene.

Domestic accelerators like Huawei Ascend and Cambricon lag Nvidia in performance per watt, suggesting a potentially lower usable training compute per gigawatt of Chinese silicon.

The facility has begun partial operations and is intended to support Z.AI’s GLM systems, with power capacity enough to power about 750,000 homes.