Z.ai, the Chinese AI lab formerly known as Zhipu, has completed construction of a 1 gigawatt class data center built entirely on Chinese made chips and has begun partially operating it, Bloomberg reported. It is the first reported case of a Chinese big tech firm building a facility of this scale exclusively on domestic silicon.
The scale is substantial by any standard. One gigawatt is roughly the electricity needed to power 750,000 homes at any given moment, and the hub is designed to train Z.ai's GLM family of frontier models. The company already operates several computing clusters with more than 10,000 chips each.
The harder problem than buying domestic chips is making them work together. Running one model efficiently across accelerators from different vendors, such as Huawei and Cambricon, requires a software layer that most labs get for free from Nvidia's ecosystem and Chinese labs have had to build themselves. Z.ai addressed that directly, completing the acquisition of Zhongke Jiahe, an infrastructure software company founded by researchers from the Chinese Academy of Sciences whose technology runs the same AI model efficiently across different Chinese made chips.
The business behind the build out is growing fast. Z.ai says it expects to reach 1 billion dollars in annual recurring revenue, having already hit its 2026 sales target in July, and its GLM models are among the most widely adopted open weights in the world.
United States export controls were designed to slow Chinese AI development by limiting access to Nvidia silicon. The answer taking shape is substitution at scale. Domestic accelerators still trail Nvidia chip for chip, but a gigawatt of them, tied together by cross chip software, is a capacity answer, and it arrives in the same week that serving capacity, not model quality, became the defining bottleneck of the open model wave.