Z.AI Launches 1 Gigawatt Data Center. The Goal: Prove That China Does Not Need NVIDIA

Z.AI, previously known as Zhipu and one of China’s leading providers of agent-based AI specialized in coding, has partially launched a gigantic 1 gigawatt data center for artificial intelligence. The uniqueness of the project lies in the hardware: according to reports from Bloomberg, the infrastructure exclusively uses chips produced in China, without relying on NVIDIA accelerators.

The site is intended for training the GLM models developed by Z.AI and represents one of the largest server hubs ever created by a Chinese AI laboratory. To provide a point of comparison, a power of 1 GW is approximately equivalent to the simultaneous consumption of 750,000 households.

According to the source cited by Bloomberg, Z.AI operates multiple computing clusters, each composed of over 10,000 chips. The exact models of the accelerators used have not been disclosed, but among the main Chinese AI chip producers are Huawei, Cambricon, and Alibaba.

This data is particularly significant because the export controls imposed by the United States limit Chinese companies' access to the most advanced NVIDIA GPUs. So far, the main doubt was regarding the capability of domestic chips to support not only the execution of models but also the training of cutting-edge AI systems.

The establishment of a 1 GW cluster entirely based on domestic chips provides a concrete answer to this question. The project suggests that Chinese laboratories can continue to increase their computing capacity even in the absence of GPUs that are considered standard in the rest of the industry.

The timing of the announcement is interesting. A few days prior, the Beijing-based company Moonshot had presented the Kimi K3 model, capable of competing with some of the best American models, but had to temporarily suspend new registrations due to a shortage of computing capacity.

In contrast, Z.AI aims to directly control the hardware infrastructure, thereby eliminating dependence on external suppliers. Meanwhile, China is preparing investments of around 2 trillion yuan over the next five years for the construction of data centers across the country.

After its listing in Hong Kong and a subsequent stock sale, Z.AI expects to achieve 1 billion dollars in annual recurring revenues, thus becoming the first Chinese supplier of AI models to reach this threshold. The company’s shares saw an increase of almost 20% on the day following the news.

However, some uncertainties remain. Z.AI has not officially confirmed details regarding the chips used, and the construction of the cluster does not automatically guarantee that the GLM models can be trained with the same efficiency as infrastructures based on NVIDIA GPUs.