Z.ai shares surge 37% after 1-GW Chinese-chip data centre launch
Synopsis
Key Takeaways
Z.ai, the Beijing-based artificial intelligence company also known as Zhipu, saw its shares soar 37 per cent on the Hong Kong stock exchange on Tuesday, 21 July 2026, closing at HK$1,219 (US$155), after the firm completed construction of a massive AI computing facility powered entirely by Chinese chips.
The data centre at the centre of it all
According to people familiar with the matter, Z.ai has built a 1-gigawatt AI computing centre — one of the largest of its kind — designed to train and deploy its proprietary GLM family of large language models. The company has positioned its flagship GLM-5.2 as one of China's leading large language models (LLMs). The facility's reliance on domestically produced chips marks a significant operational milestone as Chinese AI firms navigate tightening United States export restrictions.
Why it matters
The stock rebound is particularly striking given that Z.ai shares had shed more than 40 per cent over the preceding week. The 37 per cent single-session recovery signals that investors view the data centre completion and the domestic chip strategy as a credible hedge against supply-chain vulnerabilities. With Z.ai on the United States Entity List, sourcing compute infrastructure entirely from Chinese silicon is no longer optional — it is existential.
XCore Sigma acquisition adds software muscle
Alongside the data centre announcement, Z.ai has also completed the acquisition of XCore Sigma, a Chinese infrastructure software developer spun off from the Chinese Academy of Sciences, according to people familiar with the matter. XCore Sigma specialises in heterogeneous computing software — including compilers, runtime systems, and inference engines — that improve chip utilisation across vendors, lower inference costs, and accelerate model deployment. The combined move is intended to address computing power constraints and sharpen Z.ai's AI inference efficiency.
The competitive backdrop
Z.ai's dual push — sovereign compute infrastructure plus indigenous software optimisation — mirrors a broader race among Chinese AI labs to build full-stack, self-reliant technology pipelines. Rivals including Baidu, Alibaba, and Huawei's AI division have similarly accelerated domestic chip integration following successive rounds of US export controls. The XCore Sigma acquisition gives Z.ai a software layer that can squeeze more performance from heterogeneous chip arrays — a critical advantage when access to leading-edge US-designed GPUs remains restricted. Z.ai did not immediately respond to a request for comment.
What's next
The key question now is whether the 1-GW facility can deliver training throughput competitive with frontier labs operating on Nvidia hardware. Analysts and investors will be watching GLM-5.2 benchmark results and inference cost metrics in the coming quarters as the clearest test of whether domestic silicon can close the gap. The XCore Sigma integration timeline will also be a bellwether for how quickly Z.ai can operationalise its new software stack at scale.