Envision’s Galaxy AI campus in Ulanqab, Inner Mongolia, has two wind farms of its own with a combined 420MW of installed capacity, about 1.75 times the 240MW datacenter load of its first two phases, according to a site report published by SemiAnalysis. The report says Galaxy “relies largely on wind”.
SemiAnalysis says Envision aims to supply over 60% of the first two phases’ power use with its own wind farms. By our arithmetic, 420MW running for 3,500 full-load hours, the report’s figure for a good Ulanqab site, gives about 1.47 TWh a year, roughly 70% of the 2.10 TWh the 240MW load would use running flat out. That is gross annual energy, not hour-by-hour matching.
An evergreen Business Engineer explainer on grid cost allocation: who pays for new power lines. It teaches the general idea only and does not discuss Galaxy or Envision.
The key insight: SemiAnalysis says Envision’s Galaxy campus in Ulanqab has two wind farms with 420MW of installed capacity for a 240MW load in its first two phases, about 1.75x, and that Envision aims to supply over 60% of that power use with its own wind. By our arithmetic, the farms’ gross annual output equals about 70% of the load’s full-use demand.
SemiAnalysis says Galaxy’s power comes through a model China calls source-grid-load-storage, or SGLS. Under it, “a project builds its own wind and/or solar farm, battery storage and datacenter, and links them through a dedicated line instead of the public grid”.
For hyperscalers, the report says, the appeal of an SGLS project is “long-term power at a largely fixed cost”, and power from wind and solar also counts toward green targets. SemiAnalysis says all four phases of Galaxy are built-to-suit for two internet giants.
The setting matters. SemiAnalysis says Inner Mongolia holds about 57% of China’s usable wind resources, with wind farms there averaging about 2,800 full-load hours a year, a capacity factor of roughly 32%, while Ulanqab is windier at 35–40%. It adds that policy encourages bringing new demand to where the renewable resources are, and puts power in Inner Mongolia at about $0.055/kWh all-in, “among the lowest in the world”.
Galaxy, Ulanqab: two wind farms with 420MW of installed capacity against a 240MW datacenter load in the first two phases, about 1.75x. Data: SemiAnalysis, free section. Our arithmetic: 420MW x 3,500 full-load hours = about 1.47 TWh a year vs 240MW x 8,760 hours = 2.10 TWh, roughly 70%, gross annual energy, not hourly matching.
Why Oversize the Wind
The report lists three factors behind the 60% aim. On siting, Envision chose windy sites about two hours’ drive from the campus, and the two wind farms sit about 80km apart, which SemiAnalysis says cuts the hours in a year when neither has wind. On scale, the farms total 420MW against the 240MW load. On storage, a 4-hour battery system shifts power from windy hours into calm ones.
SemiAnalysis illustrates the trade-off with stylised curves. With 1MW of wind for every 1MW of load, wind covers 40% of the year’s power. Doubling the wind farm lifts wind’s share to about 55%, the report says, but in windy hours the farm then produces more than the campus can use, and that power must be stored or curtailed.
Envision has run the idea before. SemiAnalysis says that in Chifeng it built a 2GW off-grid wind, solar and storage system that feeds a green ammonia plant, running entirely on power electronics with no synchronous generator, and that a small pilot datacenter built with Tencent’s modular T-Block units runs on the same system. The report calls Galaxy “a larger application of the same idea”.
Two Campuses Planned Around Their Own Power
SemiAnalysis compares Galaxy with Stargate’s flagship campus in Abilene, Texas, and says both “were planned around their own power”. In its words, “Abilene draws on the grid and on gas turbines on site, while Galaxy relies largely on wind”. We reported the headline facts on the campus in China AI Campus Built 120 MW in a Year at a Fifth of US Cost.
As we read the two cases, power is being secured through equipment. In the US case reported by the FT, that means gas generators ordered years earlier. In Envision’s case, as SemiAnalysis describes it, the integrator that makes the turbines, batteries and power electronics also builds the load.
SemiAnalysis writes that “For Envision, datacenters create new demand for the power and equipment it already produces”. It also says project returns across China’s datacenter buildout are thin, “with IRRs squeezed hard”, so the reward often comes from elsewhere. The report says Envision’s datacenter plans in Inner Mongolia add up to 3GW, with Galaxy the first.
The Structural Read
As we read the report, oversizing trades surplus for coverage: SemiAnalysis’s stylised curves show wind’s share rising from 40% to about 55% when the farm is doubled, with the extra output in windy hours stored or curtailed.
As we read SemiAnalysis’s comparison with Abilene, the two campuses share a starting point, their own power, but differ in the source: the grid and on-site gas turbines in Texas, largely wind in Ulanqab.
As we read the Chifeng project, Envision has already run a small pilot datacenter on an off-grid wind, solar and storage system built with Tencent’s modular T-Block units.
SemiAnalysis, We Visited China’s Stargate in Inner Mongolia
“For hyperscalers, the appeal of an SGLS project is long-term power at a largely fixed cost”
Three Implications
FOR BUYERS OF COMPUTE SemiAnalysis says the appeal of an SGLS project for hyperscalers is long-term power at a largely fixed cost, and that power from wind and solar also counts toward green targets.
FOR EQUIPMENT MAKERS SemiAnalysis says that for Envision, datacenters create new demand for the power and equipment it already produces.
FOR THE NUMBERS TO WATCH SemiAnalysis says Envision’s datacenter plans in Inner Mongolia add up to 3GW and that the full Galaxy campus is expected to be completed by the end of 2027.
The essay says “The structural bottleneck isn’t computing, it’s electrons.”
As we read it, Galaxy’s own wind farms and the gas generators in the US case are two ways of buying those electrons as equipment. The essay does not discuss Galaxy or Envision.
What Is Not Established
The 70% figure is our gross annual arithmetic, using the report’s 3,500 full-load hours for a good Ulanqab site and assuming the campus runs flat out; it is not hourly matching and ignores curtailment and storage losses. The share Envision states is an aim of over 60%, not a measured result. The report’s construction and design detail sits behind a paywall that we did not read.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
According to SemiAnalysis, Envision’s Galaxy campus pairs 420MW of its own wind with a 240MW load in its first two phases and aims to cover over 60% of the first two phases’ power use with that wind. By our arithmetic the wind farms’ gross annual output equals about 70% of the load’s full-use demand. SemiAnalysis says the Abilene campus in Texas draws on the grid and on-site gas turbines instead.
A note on sourcing. The figures and quotations on Galaxy, Envision and Inner Mongolia are from the free section of SemiAnalysis’s report We Visited China’s Stargate in Inner Mongolia, read on 8 November 2026; the paywalled part was not read. The Anthropic and Caterpillar figures are from our earlier reporting on the Financial Times story. The 1.47 TWh, 2.10 TWh and 70% figures are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
