The global AI compute migration: who is the next Ulanqab?
A 36Kr analysis traces how AI compute is shifting from tech hubs toward energy-producing regions: as of June, Ulanqab in Inner Mongolia had secured about 12.5GW in data-center capacity commitments, exceeding the 10GW initial US commitment OpenAI announced when it launched Stargate in 2025. The piece cites local data-center power prices of roughly RMB 0.32–0.36 per kWh, close to ten months a year of natural cooling with PUE as low as about 1.1, and more than 75,000P of computing demand served for Beijing customers. It lists Aragon in Spain (Amazon's EUR 33.7bn cumulative investment), Johor in Malaysia, Chiba in Japan (JERA with Dell Technologies and RHAELM, over $15bn), Kazakhstan and Uzbekistan, and the UAE as contenders, noting the UAE shifted toward distributed and underground facilities after regional conflict. It also mentions Australian AI infrastructure firm Firmus signing a multiyear deal to supply OpenAI with computing from its Malaysia sites.
As competition shifts from scale to a mix of power price, climate, latency and permitting speed, site selection is becoming a new gating factor for AI infrastructure.