Chinese AI Firms Pledged 12.5 GW Data Center Expansion in Ulanqab
Inner Mongolia city faces water rationing as tech giants rush to build gigawatt-scale compute clusters.
Chinese technology companies have pledged to construct 12.5 gigawatts of data center capacity in Ulanqab, Inner Mongolia, with over 70 percent announced in the past year. Driven by cheap electricity and cold weather, the rapid buildout is straining local water supplies in the arid region, forcing municipal water utilities to enact nightly supply cutoffs.
Key details
According to analysis by Goldman Sachs published in August 2026, Chinese AI firms and cloud operators have opened or begun construction on nearly 100 data centers in Ulanqab since 2016. The city, located on the Inner Mongolian Plateau about two hours west of Beijing, has seen commitments surge as domestic AI developers like DeepSeek, ByteDance, Alibaba, and Xiaohongshu shift toward building proprietary compute infrastructure. For comparison, OpenAI's planned $500 billion Stargate project targets 10 gigawatts of total capacity.
The expansion is favored due to low electricity rates and natural cold weather that reduces cooling needs for much of the year. However, Ulanqab receives only roughly 14 inches of annual rainfall, comparable to arid climates like Denver. To manage peak municipal demand during summer cooling periods, the local water utility was forced to shut off select waterworks for seven hours each night. Additionally, while the region is expanding wind and solar capacity—such as Envision's planned 2-gigawatt clean power-linked data center—coal still supplies approximately 37 percent of Ulanqab's electricity grid to maintain continuous operations.
Why this matters
The concentration of 12.5 gigawatts of proposed compute capacity underscores the colossal resource demands of scaling frontier AI models in Asia. Unlike previous cloud infrastructure used primarily for cold storage, domestic AI developers are deploying high-density clusters in Inner Mongolia for both model training and low-latency inference. The resulting water scarcity highlights how high-density AI cooling requirements directly conflict with local municipal and agricultural water security in fragile ecological regions.
Context
China's national "Eastern Data, Western Compute" initiative launched in 2021 initially aimed to shift data workloads to western provinces. Historically, high latency limited these remote facilities to data backups. Dedicated fiber-optic links completed in 2017 and 2019 reduced latency between Ulanqab and Beijing to under 5 milliseconds, making the region viable for real-time AI inference. While Inner Mongolia is rapidly scaling renewable generation, its reliance on baseload coal highlights the ongoing challenge of decarbonizing continuous AI workloads.
Risks and open questions
A primary risk is whether Ulanqab's limited water reserves can sustain 12.5 gigawatts of compute capacity without triggering permanent municipal water crises or severe agricultural disruptions. Although high-altitude winters allow data centers to operate with reduced cooling water for ten months of the year, summer thermal stress remains a major operational bottleneck. Furthermore, it remains uncertain how quickly operators can transition away from coal power without compromising grid reliability for high-density AI clusters.
What happens next
Local authorities and data center operators are expected to implement mandatory dry-cooling retrofits and closed-loop liquid systems to cut operational water draw during summer peak periods. Municipal regulators will also evaluate power access agreements to align new gigawatt-scale construction with direct renewable energy installations like Envision's 2-gigawatt off-grid project.
Source: WIRED Published on AI Usage Global, author: AUG Bot
