In a powerful demonstration of China’s growing self-sufficiency in artificial intelligence infrastructure, Beijing-based developer Z.ai has completed the construction of a massive 1-gigawatt AI data center built entirely using domestically produced chips. The facility, which is already partially operational, represents one of the largest data centers ever constructed by a Chinese AI lab and serves as a direct rebuke to US efforts to constrain China’s AI development through semiconductor export controls.
According to a Bloomberg report cited by Technode, the new data center is designed to support the training and development of Z.ai’s frontier GLM model family. The scale of the facility is staggering; a 1-gigawatt power capacity is roughly equivalent to the electricity required to run 750,000 homes. While the specific location, cost, and construction timeline of the data center have not been disclosed, its completion marks a significant milestone in China’s push to build indigenous computing infrastructure.
Overcoming the Nvidia Embargo
Z.ai, formerly known as Zhipu AI, open-sourced its GLM-5.2 model as a direct counter to US restrictions and has been on the US Commerce Department’s Entity List since January 2025, legally barring the company from acquiring advanced AI accelerators from American firms like Nvidia. Forced to rely entirely on domestic hardware, Z.ai has aggressively scaled its operations using Chinese alternatives. While the specific chip supplier for the new 1-gigawatt facility was not named in the reports, it is notable that the company’s GLM-5.2 model, released in June, was trained entirely on Huawei’s Ascend accelerators.
The successful deployment of a data center of this magnitude using domestic chips proves that Chinese hardware, despite trailing Nvidia’s latest Blackwell architecture in performance per watt, can be effectively scaled to meet the demands of frontier AI training. By networking massive numbers of less efficient chips together, companies like Z.ai are brute-forcing their way past the limitations imposed by US sanctions. The company also reportedly operates several other computing clusters, each containing more than 10,000 chips.
A National Infrastructure Push
The Z.ai facility aligns with a broader national strategy to build a robust, self-reliant AI infrastructure. Beijing is currently drafting a plan to spend approximately 2 trillion yuan ($295 billion) over the next five years to develop a national AI data center grid, with a mandate that 80% of the chips used must be domestically produced. This massive state-backed investment is driving rapid expansion across the domestic semiconductor supply chain.
Reports indicate that China’s leading foundry, SMIC, is running its advanced N+2 manufacturing process at above 93% utilization to meet surging domestic demand for AI chips. Huawei alone shipped an estimated 812,000 AI chips last year. For Z.ai, the investment in domestic infrastructure is paying off commercially; the company recently raised billions through a Hong Kong IPO and follow-on share sale, and is reportedly on track to reach $1 billion in annual recurring revenue after hitting its 2026 sales target in July.
GLM Models and the Enterprise AI Push
The 1-gigawatt data center is the computational backbone of Z.ai’s ambition to become China’s leading enterprise AI platform. The company’s GLM model family, which includes GLM-5.2, trained entirely on Huawei Ascend chips, powers a range of enterprise applications including code generation, document analysis, and multimodal reasoning. By controlling its own computing infrastructure, Z.ai can iterate on its models faster and offer enterprise clients guaranteed performance levels that would be difficult to promise if the company were dependent on third-party cloud providers.
The data center also positions Z.ai favorably in the context of Beijing’s push for data sovereignty. Chinese enterprises and government agencies are increasingly required to process sensitive data on domestically owned and operated infrastructure. By building a massive, wholly domestic computing facility, Z.ai can credibly offer its enterprise clients the assurance that their data never leaves Chinese-controlled systems, a significant competitive advantage over foreign cloud providers and even some domestic rivals who rely on foreign hardware. As China’s AI industry matures and enterprise adoption accelerates, companies that control the infrastructure layer will be well positioned to capture a disproportionate share of the value being created.
The 1-gigawatt milestone also carries symbolic weight in the broader US-China technology competition. When the US imposed export controls on advanced Nvidia chips, the explicit goal was to prevent China from building the computing infrastructure necessary to train frontier AI models. Z.ai’s data center, built entirely on domestic chips and already operational, demonstrates that this goal has not been achieved.
It is a concrete, measurable rebuttal to the premise that export controls can effectively constrain China’s AI development and a signal to policymakers in Washington that the window for using chip restrictions as a decisive strategic lever may be closing faster than anticipated. For China’s AI industry, the Z.ai facility is a proof-of-concept: that the combination of domestic chip production, massive state investment, and determined engineering can overcome the hardware constraints imposed by geopolitical rivalry, at least at the scale required to train and deploy competitive frontier models.