Z.AI reportedly completes a gigawatt data center built entirely on Chinese chips

China’s Z.AI has completed construction of a data center filled entirely with Chinese-made chips, part of Beijing’s drive to cut the country’s reliance on restricted Nvidia silicon, according to reporting published July 20, 2026.

Z.AI, the company formerly known as Zhipu, has begun partially operating the roughly one-gigawatt facility, enough electricity to power about 750,000 homes, to develop its GLM family of AI models, the report said. Z.AI now runs several computing clusters that each hold more than 10,000 chips.

The significance is less the raw scale than the supply chain. U.S. export controls have choked off China’s access to Nvidia’s most capable AI accelerators, and a gigawatt-class training site running only on domestic parts would be evidence that Chinese chipmakers can now field enough silicon to train frontier models without American hardware, the outcome the controls were meant to prevent.

Big caveats apply. The report does not specify which domestic chips fill the site, how they perform against Nvidia’s, or how much of the gigawatt is actually energized, and Z.AI has not publicly confirmed the project. Power capacity is also not the same as usable compute: cooling, networking and chip yields determine how much of that gigawatt turns into model training. A facility built on Chinese accelerators may still lag the efficiency of Nvidia-based clusters abroad.

Z.AI is also behind GLM-5.2, the open-weight model that a separate incident this month showed being used for security work outside China. Whether its domestic-chip data center delivers competitive training runs, or mainly demonstrates political self-sufficiency, will be clearer as the company details what the GLM models trained there can do.