Chinese AI companies are building nearly 100 data centers in Ulanqab, Inner Mongolia, with 12.5 gigawatts of pledged capacity.
The region offers cheap electricity and cold winters but faces acute water scarcity and reliance on coal power.
This marks the first time Chinese AI firms are building their own infrastructure instead of renting from cloud companies.
What happened
Chinese AI companies including DeepSeek, ByteDance, and Alibaba are building nearly 100 data centers in Ulanqab, Inner Mongolia, with combined pledged capacity of 12.5 gigawatts—over 70 percent announced in the last year alone, making it one of Asia's fastest-growing compute clusters.
Why it matters
For the first time, Chinese AI companies are investing heavily in their own infrastructure rather than renting from cloud providers, signaling they are finally catching up to American peers in physical compute spending. The shift became viable once AI model training proved less sensitive to latency than earlier backup-storage uses.
What to watch
Ulanqab receives only roughly 14 inches of rain per year and already struggles to meet resident water demand; data centers require additional cooling water year-round, though less in winter, posing a significant environmental challenge. About 37 percent of Ulanqab's electricity still comes from coal despite government plans to transition to renewables.
Ask the AI about this article →
Ulanqab's emergence as China's premier AI data center hub reflects a fundamental shift in how Chinese AI companies approach infrastructure. For years, firms like DeepSeek and ByteDance relied on renting compute from cloud providers, spending far less on physical infrastructure than American counterparts. The article indicates this changed as two conditions aligned: first, AI model training's low sensitivity to latency made remote, low-cost locations viable in a way backup storage never did; and second, Chinese AI startups began attracting enough paying users to justify the capital investment in dedicated inference capacity. The location itself is not accidental. Inner Mongolia had been designated in 2021 as a hub under the government's "Eastern Data, Western Compute" initiative, but initial drawbacks—remote location and high latency to coastal population centers—limited adoption until fiber optic cables built in 2017 and 2019 reduced latency below five milliseconds, enabling real-time inference workloads.
The environmental tension underlying this boom is acute. While the central government views data centers as a tool to absorb China's excess renewable energy capacity—creating a narrative of win-win economic and green progress—the reality is more complicated. Research cited in the article shows 37 percent of Ulanqab's electricity still derives from coal, not renewables. Data centers running 24/7 historically favor fossil fuels for reliability. Water scarcity is the more immediate crisis: Ulanqab receives only roughly 14 inches of rain annually and the local water company is already unable to meet current resident demand, having cut supply for seven hours nightly last month. Although data centers require less cooling water in winter (two months of the year), the cumulative impact of nearly 100 facilities poses a significant environmental challenge. One expert quoted in the article suggests the transition to fully renewable power might take three years, implying the coal dependency is structural in the near term.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
