In China, up to 80% of newly constructed data center space is underutilized as GPU rental prices decline.

The US behemoths are not the only ones investing heavily in AI infrastructure. Last week, Alibaba Chairman Joe Tsai made news when he cautioned against an AI bubble arising from US intentions to construct AI data centers without any indication of future demand. Although a large portion of it has been useless, Chinese companies are also contributing billions to data centers.

China's government-driven data center development boom began when ChatGPT arrived. Although over 500 projects at the local and provincial levels have been launched, investment has stopped as the economy collapses.

According to an MIT Review, 80% of newly constructed capacity is reportedly sitting idle, and hundreds of new data centers nationwide are hardly operating at all. One contributing element is the DeepSeek revolution, which has upended the cost and investment dynamics surrounding China's AI manufacturing and raced across the business sector. Despite the exorbitant price of purchasing Nvidia GPUs in China, it has also contributed to a decline in GPU leasing costs.

According to reports, the monthly leasing fee for an Nvidia H100 server with eight GPUs has decreased from a peak of over RMB180,000 ($24,802) to 75,000 Chinese yuan (US$10,317). Businesses are attempting to sell off GPUs and abandoning data center investments. Given the significant energy costs associated with maintaining a limited number of Nvidia servers, some people believe it is more cost-effective to leave data centers empty.