While the chip supplier was not identified, Z.ai's recent development efforts have centered on Huawei hardware. The company launched its GLM-5.2 open-weight model in June, saying it was trained entirely on Huawei Ascend AI accelerators without using Nvidia chips. Z.ai has been unable to legally access Nvidia hardware since being added to the U.S. Commerce Department's Entity List in January 2025, making domestically produced AI chips its primary option for large-scale model training.
The 1-gigawatt facility highlights the scale of China's investment in AI infrastructure, though electricity capacity alone does not determine computing performance. Chinese AI accelerators, including Huawei's Ascend lineup, currently trail Nvidia's latest Blackwell chips in performance per watt, meaning a similarly sized power footprint delivers less training compute than an equivalent Nvidia-based installation.
The project comes as China advances broader plans to expand domestic AI infrastructure. Beijing is drafting a proposal to invest about 2 trillion yuan, or approximately $295 billion, over five years to develop a nationwide network of AI data centers, with at least 80% of the underlying technology sourced from Chinese suppliers.
However, manufacturing enough AI hardware to fill those facilities remains a challenge. The country's most advanced stable semiconductor process from SMIC is operating above 93% utilization, while limited domestic supplies of high-bandwidth memory constrain production of Huawei's AI accelerators. Huawei shipped about 812,000 AI chips last year, underscoring the gap between infrastructure ambitions and available hardware.
The announcement also comes as demand for AI computing continues to rise across China's AI sector. Rival startup Moonshot recently paused new subscriptions following the release of its Kimi K3 model, saying it wanted to preserve computing resources for existing users.
According to Bloomberg, Z.ai is on track to reach $1 billion in annual recurring revenue after meeting its 2026 sales target in July. The company has also raised billions of dollars through a Hong Kong initial public offering and a subsequent share sale.
Bloomberg's source did not disclose the location of the new data center, how much it cost to build, or how long construction took.
This analysis is based on reporting from Tom's Hardware.
Image courtesy of Unite.AI.
This article was generated with AI assistance and reviewed for accuracy and quality.
