AI: a warning from China

China’s AI boom hopes crash: More than 80% of newly built AI data centres remain unused

Post-pandemic, China invested heavily in artificial intelligence (“AI”) infrastructure, building hundreds of data centres to capitalise on the AI boom. But many of these facilities now stand unused due to weak demand and a shift in AI trends driven by DeepSeek.

The post-pandemic economic downturn and the slump in the real estate sector led Chinese officials to scramble for alternative growth drivers. AI infrastructure became the new stimulus of choice.

But the rush to build AI data centres was driven from the top down, often with little regard for actual demand or technical feasibility and many projects were led by executives and investors with limited expertise in AI infrastructure, resulting in hastily constructed centres that fell short of industry standards.

China’s data centre gold rush was fuelled by government and private investors, with over 500 new data centre projects announced in 2023 and 2024; at least 150 of them were completed and operational by the end of 2024.

The demand for high-performing Nvidia graphics processing units (“GPUs”), which were essential for training AI models, was extremely high at one point, with prices reaching up to 200,000 yuan ($28,000) on the black market. But prices have since decreased and many traders are now selling off their surplus GPUs due to poor returns.

“Renting out GPUs to companies that need them for training AI models – the main business model for the new wave of data centres – was once seen as a sure bet. But with the rise of DeepSeek and a sudden change in the economics around AI, the industry is faltering,” MIT Technology Review said.

According to reports from local Chinese media outlets, up to 80% of China’s newly built computing resources remain unused. Most companies running these data centres are struggling to stay afloat, with some projects failing.

The situation has prompted concerns about the waste of energy and resources, and the Chinese government may eventually step in to take over and hand off the distressed assets to more capable operators as the industry continues to falter and the economics around AI change.

China’s efforts to catch the AI boom by building hundreds of AI data centres have resulted in many of these centres standing unused, with only about 10% of companies still actively investing in large-scale model training by the end of 2024.

Some brokers were not looking to make money from data centres in the first place but were instead interested in exploiting government benefits, such as subsidised green electricity and state-backed loans and credits, with some companies acquiring land for data centre development to qualify for these benefits and then leaving the facilities unused.

Despite the underutilisation of data centres, China’s central government is still supporting the development of AI infrastructure. In early 2025, it convened an AI industry symposium, emphasising the importance of self-reliance in this technology. Major Chinese technology companies, such as Alibaba Group and ByteDance, are also making significant investments in cloud computing and AI hardware infrastructure. Alibaba is planning to invest over $50 billion over the next three years, and ByteDance is planning to invest around $20 billion in GPUs and data centres.

However, Fang Cunbao, a data centre project manager based in Beijing, believes that the industry has changed and that no clear-headed contractor or broker would still expect to make a direct profit from data centres. “Everyone I met is leveraging the data centre deal for something else the government could offer,” Fang said.

Jimmy Goodrich, senior advisor for technology at the RAND Corporation, attributes the growing pains in China’s AI industry to inexperienced players, including corporations and local governments, jumping into the market and building facilities that are not optimal for current needs, leading to distressed assets and potential government intervention.

According to Goodrich, China is unlikely to scale back its efforts in developing AI infrastructure, as the country views underused data centres as a necessary step towards developing an important capability, and the state will likely consolidate and clean up failed projects and distressed assets.

The above is paraphrased from the article ‘China built hundreds of AI data centres to catch the AI boom. Now many stand unused’ written by Caiwei Chen and published by MIT Technology Review on 26 March 2025. Read the full article HERE.