- Deepseek is just one of the many reasons why China’s growth simply did not materialize
- Up to 80% of the capacity of a new data center has not been used according to local sources
- If this capacity reaches the widest market, it could cause a great headache to data centers developers
The rise of China’s infrastructure is hesitant, since according to a report in MIT technology reviewThe country built hundreds of data centers to support their AI ambitions, but many are not used now.
The state and private entities invested billions in 2023 and 2024, with the expectation that the demand for GPU rentals would continue to grow, but the absorption has decreased and, as a result, many operators are now struggling to survive.
Much of the early impulse was driven by exaggeration. The government, anxious for China to become a world leader at AI, encouraged local officials to the construction of accelerated data centers with the result that more than 500 projects were announced throughout the country, and at least 150 were completed at the end of 2024, according to sources affiliated with the State. However, MIT technology review He says that local publications report that up to 80% of this new computer capacity remains inactive.
Sell GPU
The location is also a problem, MIT technology review grades. Facilities built in the center and western China, where electricity is cheap, now face the problems that meet the latency requirements. According to reports, in cities like Zhengzhou, operators even give free computer coupons in an attempt to attract users.
In some regions, developers began selling GPU after not getting long -term customers.
Xiao Li, a project manager of the Data Center who spoke with MIT technology reviewHe said that many Wechat groups that were once boasted from Nvidia chips agreements have been silent. “It seems that everyone is selling, but few are buying,” he said.
If this capacity reaches the broader market, it could cause a great headache to data centers developers, flooding a sector already soft with even more supply and pushing prices even more.
A reason for the fall in demand is the emergence of Deepseek, which overturned the global technological economy when it was launched in January 2025.
Its open source reasoning model, R1, coincides with the performance of Chatgpt O1 but at a lower cost, displacing the interest of the training of the model and towards inference: the real -time use of AI models, which requires a different infrastructure.
Many of the data centers built during the race were designed for large -scale training, not the demands of low latency of real -time reasoning.
Despite the excess offer, the Chinese government remains committed. The central authorities celebrated an AI symposium at the beginning of 2025, and companies such as Alibaba and Tiktok owner bya have announced large investments.
However, for many early data centers, expectations have collapsed. The infrastructure was built, but the demand that was destined to supply simply has not arrived.
