Regions across Inner Mongolia, including Ulanqab and Hohhot, are establishing themselves as central hubs for China's artificial intelligence infrastructure. These areas offer low electricity rates and a cooler climate advantageous for data center cooling. The China Telecom-Inner Mongolia Information Park ranks among the world's largest data centers.
Zhangjiakou, located in Hebei province adjacent to Beijing, plays a critical role in China's Eastern Data Western Compute (EDWC) project. Its proximity to the capital provides lower latency for cloud and AI applications, making it more attractive for private data center investment than remote clusters.
Beijing drew on Zhangjiakou for infrastructure during the 2022 Winter Olympics, including renewable power generation. Existing water infrastructure from the Guanting Reservoir, built in the 1950s, now positions Zhangjiakou to support Beijing's expanding cloud computing needs.
The EDWC project, launched to coordinate data center energy demands with renewable energy expansion, sought to prevent overbuilding in costly urban areas and concentrate facilities where cheap power was available. China's National Data Bureau reported 45.5 billion yuan, or $6 billion, of capital invested in data centers through the plan.
Yet the energy reality diverges from the green strategy. Data centers in Zhangjiakou consumed 1.859 billion kilowatt-hours of electricity in the measured period, up 43.47 percent year-over-year. Green electricity accounted for 574 million kilowatt-hours—approximately 30 percent of the total. The remainder comes from coal or gas-fired power plants.
The mismatch reveals a structural problem: facilities built near renewable sources often draw heavily on traditional grids to ensure the reliable power supply critical for continuous operations. Local permitting processes have also been compromised by bribes to secure building permits, contributing to speculative overbuilding that may outstrip actual renewable capacity.
Reports indicate that portions of newly constructed data centers remain underutilized or empty, suggesting construction has outpaced demand and operational readiness.
More remote EDWC clusters in Gansu, Ningxia, and Xinjiang have proven less attractive for most data center types. Distance from major cities creates latency challenges and makes it harder to recruit trained engineers. These remote locations could serve niche applications—such as training large AI models, which tolerate higher latency—but their utility remains limited for real-time computing tasks.
