Chinese Artificial Intelligence Centers Sit Empty, with Up to 80% of New Capacity Unused
Over the past two years, China has invested heavily in artificial intelligence infrastructure with the goal of becoming a global leader in the field. Despite its ambitious plans, it now faces an unexpected problem—according to the latest data, up to 80% of newly built AI data centers remain unused. This situation reveals the complicated relationship between state-driven technology policy and rapidly changing market demands.
Rushed Construction Driven by State Incentives
In 2023–2024, the Chinese government actively encouraged local officials to accelerate the construction of data centers as part of its strategy to become a global artificial intelligence powerhouse. The result was the announcement of more than 500 projects across the country, with at least 150 of them completed by the end of 2024. Billions of dollars in investment flowed from both public and private sources. Everyone expected demand for GPU rentals (graphics processing units essential for AI computing operations) to continue growing. However, this demand did not materialize to the anticipated extent, leading to the current situation of excess capacity. One of the key factors contributing to this problem is the fundamental mismatch between what was built and what the market actually needs:
- Many new facilities were designed primarily for training large-scale AI models, while current demand is shifting toward real-time inference.
- Data centers requiring low latency must be located near technology hubs, while many new facilities were built in remote or rural areas.
- Some locations offered cheap electricity but could not meet the latency requirements essential for modern AI tasks.
The "DeepSeek Effect" Had a Significant Impact on the Market
The launch of the R1 model by DeepSeek had a significant impact on reducing demand for traditional computing capacity. This open-source model delivers performance on par with ChatGPT but at significantly lower costs. As a result, many smaller companies abandoned plans for their own large-scale model pretraining, further reducing demand for traditional training-focused infrastructure—precisely when new capacity was entering the market. Many facilities are suffering from the consequences of poor planning. They were built without adequately considering actual market needs or adhering to technical standards. The situation was worsened by a shortage of qualified engineers, leading some operators to make costly errors or rely on intermediaries who overestimated future demand. And the economic impacts of this situation are considerable:
- Declining profits and falling GPU rental prices have left many operators in a difficult financial situation.
- Some operators are even giving away vouchers for free computing power in an effort to attract users.
- Early investors are now facing collapsed expectations while their infrastructure remains unused.
Government Response and Future Prospects
Despite these obstacles, central authorities remain determined to continue developing national artificial intelligence capacity. Major companies such as Alibaba and ByteDance have announced continued investments amounting to tens of billions of dollars annually. The government is expected to intervene by taking control of troubled centers and potentially reallocating them. However, the industry recognizes that the original bubble has already burst, and significant adjustments to investment strategies and expectations are underway. Data centers in remote areas, built because of low energy prices and government incentives, cannot compete with strategically located facilities near technology hubs. These remote locations cannot provide the low latency required by modern AI applications, making them less attractive to potential clients. Overall, China’s aggressive policy of building AI data centers has not only led to massive underutilization but has also exposed critical mismatches between policy-driven supply growth and evolving technological realities within global artificial intelligence markets. While China continues its efforts to become a leader in AI, it will need to address these structural problems to align its ambitions more effectively with practical market needs.
