China’s once-booming AI data center industry is facing a harsh reality check, as an oversupply of infrastructure and shifting technological demands leave many facilities underutilized and struggling to attract investment.

Following the debut of ChatGPT in late 2022, the country’s AI sector saw a surge of investment and development, particularly in data centers built to support the growing need for advanced AI model training. However, recent market changes—including new AI technologies, a drop in GPU rental prices, and hasty overconstruction—have created a crisis of oversupply.

From Frenzy to Oversupply

Xiao Li, a former real estate contractor who pivoted to AI infrastructure in 2023, has seen firsthand how quickly the market has cooled. Last year, amid intense demand and U.S. export restrictions, Nvidia GPUs—critical for training AI models—fetched as much as 200,000 yuan ($28,000) on the black market. Today, prices have dropped, and GPU traders are struggling to find buyers.

“Everyone seems to be selling, but there aren’t many buyers,” Li told MIT Technology Review.

The downturn has also affected large-scale data center projects. Li knows of two major facilities that are now struggling to secure further funding, forcing their owners to offload excess GPU stock.

DeepSeek Disrupts the Landscape

A significant factor driving this shift is the emergence of DeepSeek, a Chinese company that has developed an open-source reasoning model matching ChatGPT’s performance but at a fraction of the cost.

This breakthrough has reshaped industry priorities, shifting the focus from building large AI models to practical applications. “The burning question shifted from ‘Who can make the best large language model?’ to ‘Who can use them better?’” said Hancheng Cao, an assistant professor at Emory University.

Many rural and remote data centers, optimized for large-scale AI training, are ill-suited for the low-latency needs of real-time inference—the process that powers reasoning models. This technological mismatch is making facilities in regions like Inner Mongolia and central China less appealing to AI firms.

A Market in Crisis

The consequences of this oversupply are becoming increasingly clear. GPU rental prices have plummeted—an Nvidia H100 server with eight GPUs now rents for 75,000 yuan per month ($10,345), down from 180,000 yuan ($25,141) at the market’s peak. Some operators are even choosing to leave their data centers idle rather than operate at a loss.

Jimmy Goodrich, senior technology advisor to the RAND Corporation, attributes these struggles to inexperienced players—including corporations and local governments—who jumped on the AI boom without understanding the technical and market realities.

“The growing pain China’s AI industry is going through is largely a result of inexperienced players… building facilities that aren’t optimal for today’s needs,” Goodrich explained.

The Role of Politics in the Data Center Boom

China’s political system has also played a significant role in the crisis. Local officials, eager to boost economic growth and advance their careers, embraced AI infrastructure as a fast-track development project. However, many of these efforts proceeded without proper feasibility studies, resulting in poorly designed facilities that fail to meet the evolving needs of AI businesses.

One example is a newly built data center in Zhengzhou. Despite offering free computing vouchers to local tech firms, the facility is struggling to attract users.

The Government’s Long-Term Vision

Despite the immediate challenges, China’s central government remains committed to AI self-reliance. In early 2025, Beijing convened an AI industry symposium emphasizing the need to build independent technological capabilities. Major tech firms like Alibaba and ByteDance have already announced significant investments in cloud computing and AI hardware.

Goodrich believes that while the current oversupply is problematic, the Chinese government sees it as part of a broader strategic goal. “They see the end, not the means,” he said, suggesting that policymakers view these underused data centers as necessary stepping stones toward a dominant AI future.

A Sector in Transition

For some industry veterans, however, the instability has prompted a career shift. Fang Cunbao, a former data center project manager, recently left the sector altogether.

“The market is too chaotic. The early adopters profited, but now it’s just people chasing policy loopholes,” Fang said, explaining his decision to transition into AI education.

As China’s AI industry navigates this period of upheaval, the long-term viability of its data center infrastructure will depend on adaptability, technological innovation, and sustained government support—a balancing act that will shape the future of AI development both within China and on the global stage.