China has invested billions of dollars in the excessive construction of data centers for artificial intelligence and processing.
It is reported that 80% of the new capacity of data centers in the country is currently unused.
The growth of artificial intelligence infrastructure in China is encountering difficulties, as many of the hundreds of data centers built to support its ambitions in this field remain idle. According to recent reports, up to 80% of the new data center capacity is not being utilized, which represents a significant challenge for developers in the sector.
During 2023 and 2024, millions of dollars were invested by both state-owned and private entities, with the expectation that the demand for GPU leasing would continue to rise. However, the reality has been the opposite, as interest in these services has decreased, leaving many operators in a precarious situation.
Much of this initial rise was driven by anticipation and government support, which encouraged local officials to accelerate the construction of data centers. More than 500 projects were announced across the country, and at least 150 were completed by the end of 2024. However, a report suggests that a considerable portion of this newly built computing capacity remains unused.
The location of some data centers also presents challenges. Those facilities erected in central and western regions of China, where electricity is cheaper, face difficulties in meeting latency requirements. In cities like Zhengzhou, some operators have had to offer free computing vouchers to attract users.
Additionally, some developers have begun selling their GPUs after failing to secure long-term contracts. A project manager from a data center noted that many WeChat groups, which previously shared offers on Nvidia chips, have now ceased to function. “It seems that everyone is selling, but few are buying,” he pointed out. If this saturated capacity finally enters the market, it could lead to an oversupply in a sector that is already facing weaknesses, which in turn could cause prices to drop even further.
One of the causes of the decline in demand is the emergence of DeepSeek, an open-source reasoning model launched in January 2025 that has disrupted the global tech landscape. This model proposes performance comparable to ChatGPT but at a lower cost, redirecting interest toward inference, which is the real-time use of AI models—an application that requires infrastructure different from that intended for large-scale training.
Despite the oversupply in the market, the Chinese government remains committed to its AI vision; a symposium on the topic was recently held, and companies like Alibaba and ByteDance have announced significant investments. However, for many of the early investors in data centers, expectations have crumbled. The infrastructure is ready, but the anticipated demand simply has not materialized.
