Shares of Chinese artificial intelligence (AI) company Zhipu surged 37 percent in a single day after it emerged that the firm has built an ultra-large AI computing centre run only on Chinese-made chips. With U.S. restrictions on exports of advanced semiconductors continuing, some see the move as accelerating technological self-reliance by securing its own AI infrastructure.

South China Morning Post reported on July 21 that Zhipu recently completed construction of a 1-gigawatt (GW) AI computing centre operated solely with Chinese-made AI chips. The facility is to be used for training and inference of the company's GLM series large language models (LLMs).

After the news was reported, Zhipu shares listed in Hong Kong surged 37 percent and ended the session at 121.9 Hong Kong dollars. The stock, which had fallen more than 40 percent over the previous week, rebounded sharply on the news that it had secured large-scale computing infrastructure.

Zhipu is also expanding its software capabilities alongside strengthening AI infrastructure. The company is reported to have recently acquired Qubilas, an infrastructure software firm spun off from the Chinese Academy of Sciences. Qubilas has developed compilers, runtime systems and inference engines to enable efficient use of AI chips from multiple manufacturers in heterogeneous computing environments.

The acquisition means Zhipu has secured both hardware and software. The new computing centre is expected to provide computing power needed for AI training, while Qubilas' technology is expected to raise utilisation of various Chinese-made AI chips, lower inference costs and improve model deployment speed.

Zhipu's moves align with a broader push by Chinese AI companies to speed up building their own ecosystems under U.S. semiconductor export controls. The U.S. government restricts exports of advanced AI chips to China, and Zhipu was included last year on the U.S. Commerce Department's Bureau of Industry and Security (BIS) trade restriction list.

At the time, Zhipu said sanctions would not have a major impact on its business and that it had secured in-house the core technology stack needed to develop LLMs. This large-scale data centre build is also seen as an extension of a strategy to reduce reliance on advanced overseas chips and strengthen its own AI infrastructure.

Demand for AI computing in China is also rising rapidly. HSBC said in a June report that China's AI computing demand would increase more than five-fold from 2025 and reach about 5 GW by 2028. It said the spread of generative AI is driving simultaneous growth in demand for large-model training and inference.

Zhipu has also made clear it will focus on securing long-term technological competitiveness rather than short-term results. Co-founder Tang Jie (탕제) said in a recent internal memo that the company would focus on research and development to prepare for the era of artificial general intelligence (AGI) rather than short-term monetisation of AI applications.

The industry sees Zhipu as moving in earnest on a strategy to build an independent AI ecosystem despite U.S. sanctions, after securing both its own computing centre and AI software technology. Improving GLM model performance and inference efficiency is expected to become an important test for the competitiveness of China's AI industry.