- Deepseek is only one of the many reasons why the growth of Chinese AI simply did not materialize
- Up to 80% of the data center capacity has not been used according to local sources
- If this capacity has struck the larger market, it could cause headache to developers of data centers
The Chinese IA infrastructure boom is failing, as according to a report in MIT review of MIT technologyThe country has built hundreds of data centers to support its AI ambitions, but many are now unused.
Billions were invested by state and private entities in 2023 and 2024, while waiting for the demand for GPU rentals would continue to grow, but absorption has in fact dropped and, therefore, many operators now have trouble surviving.
A large part of the early momentum was motivated by the media threw. The government, eager for China to become an AI world leader, encouraged local officials to accelerate the construction of the data center with the result that more than 500 projects have been announced nationally, and at least 150 were completed by the end of 2024, according to sources affiliated by the state. However, MIT review of MIT technology Said that local publications report that up to 80% of this new IT capacity remains inactive.
Sell GPUs
Location is also a problem, MIT review of MIT technology Notes. The installations built in central and western China, where electricity is cheap, now confronted with problems that meet the requirements of latency. In cities like Zhengzhou, operators have even given free calculation vouchers to try to attract users.
In some regions, the developers have started to sell GPUs after failing to secure long -term customers.
Xiao Li, a project manager of the data center that spoke with MIT review of MIT technologysaid many WeChat groups that have boasted about Nvidia flea transactions have calmed down. “It seems that everyone sells, but few buy,” he noted.
If this capacity has struck the larger market, this could cause significant headaches to developers of the data center, flooding an already soft sector with even more offer and lowering prices.
One of the reasons for the drop in demand is the rise of Deepseek, which upset the global technological economy when it was launched in January 2025.
Its open source reasoning model, R1, corresponds to the performance of Chatgpt O1 but at a lower cost, by moving the interest of model formation and towards inference – real -time use of AI models, which requires a different infrastructure.
Many data centers built during the rush were designed for large -scale training, not the low latency requirements of real -time reasoning.
Despite the excess offer, the Chinese government would have been initiated. Central authorities organized an AI symposium at the beginning of 2025, and companies like Alibaba and the owner of Tiktok Bytedance announced major investments.
For many first investors in the data center, expectations have collapsed. The infrastructure was built, but the demand it was supposed to provide has simply not arrived.
