- Published on
The Unlikely Place at the Center of China’s AI Boom
- Authors
- Name
- DeepHorus
The once-remote grasslands of Inner Mongolia have unexpectedly become the epicenter of China's artificial intelligence infrastructure boom, centered on the city of Ulanqab. Located just two hours west of Beijing, this traditionally agricultural and mining region is now home to a rapid proliferation of AI data centers. Nearly 100 facilities have been opened or started construction there since 2016, making it one of Asia's fastest-growing compute clusters.
According to a recent research note from Goldman Sachs, Chinese companies have pledged projects in Ulanqab with a combined estimated power capacity of 12.5 gigawatts. Over 70 percent of these commitments were announced in the past year alone. This scale surpasses even the planned capacity of major U.S. projects like OpenAI's Stargate Project, which aims for 10 gigawatts.
Key factors driving this migration include Ulanqab's high elevation and cold winters, which significantly reduce the energy needed for cooling. Its relative proximity to Beijing also minimizes data transmission latency. Most crucially, Inner Mongolia offers some of China's lowest electricity prices, bolstered by abundant coal and growing wind and solar energy.
A notable shift is that major Chinese AI firms are now investing directly in their own infrastructure. Companies like DeepSeek, ByteDance, Alibaba, and Xiaohongshu are reportedly building massive centers, moving away from solely renting cloud compute. This marks a significant catch-up effort in physical infrastructure investment compared to their American counterparts.
However, the boom faces a critical challenge: water scarcity. Ulanqab receives only about 14 inches of rain annually, and local authorities are already struggling to meet residential demand. Last month, the local water utility was forced to shut off waterworks nightly for seven hours to manage peak usage. While data centers require less cooling in winter, the new facilities still pose a major environmental strain.
The development aligns with China's national "Eastern Data, Western Compute" strategy launched in 2021, which promotes building data centers in western regions. Historically, latency issues relegated these sites to backup storage, but the rise of AI in 2022 changed the calculus. Professor Andrew Stokols notes that AI model training runs, which can take months and don't require real-time interaction, are far less sensitive to latency.
As a result, Ulanqab has transformed from a peripheral backup hub into a critical engine for China's AI ambitions. The rapid development highlights the global race for AI infrastructure, where access to affordable power and climate advantages can reshape geopolitical and economic landscapes. The water supply constraints, however, present a looming sustainability issue that could test the region's capacity for further growth.
