America has Stargate Abilene, the gigawatt-scale campus rising on the Texas plains for OpenAI. China has its own projects of similar size, and almost nobody outside the country has seen one up close. Now a research firm that toured one in September has reported what it found — and the speed and cost are striking.

SemiAnalysis visited Envision Group's Galaxy AI datacentre in Ulanqab, Inner Mongolia, and called the speed, scale, ultra-low cost and deep vertical integration "astonishing". The account was published on 7 November, and it offers a rare on-the-ground look at how fast China is building the infrastructure behind its AI ambitions.

A year to build, a fifth of the cost

The Galaxy campus is designed for up to 800 megawatts of IT capacity, putting it in the same league as Stargate Abilene, which has 820 megawatts of critical IT capacity on 1.2 gigawatts of power. Its first 120-megawatt phase was delivered in roughly a year, with full campus completion expected by the end of 2027. Envision, a major Chinese renewable energy company, started commercial operations at the site in August 2026 and has a longer-term target of 3 gigawatts of capacity by the end of 2028.

SemiAnalysis puts construction costs at less than $3 million per megawatt, which the authors describe as about a fifth of what a comparable site would cost in the United States. The cost comparison is the firm's own estimate, not an audited figure. The wider Chahar High-tech Zone, where the campus sits, counts 22 datacentre projects and nearly 30,000 construction workers on site, according to a local report cited by SemiAnalysis.

Wind power built in from the start

What sets the campus apart is how much of it Envision built itself. Beyond running the datacentre, the company built its own wind farms and a dedicated power line to supply it. Two wind sites roughly 80 kilometres apart carry 420 megawatts of installed capacity and supply over 60 percent of the first two phases' power, backed by a four-hour battery storage system. More than 80 percent of the site's electricity is slated to come from wind and solar in the long run.

The engineering also departs from the usual playbook. In a five-megawatt pilot hall, the campus runs 800-volt direct current all the way to the rack: solid-state transformers take 10-kilovolt alternating current from the grid and convert it straight to 800-volt DC. About a quarter of Chinese datacentres built in the past decade already run on DC, the authors note, against about 97 percent of US sites still on AC power systems.

Scale and demand

For a sense of scale, the main site covers 120,000 square metres and is designed to hold up to one million AI accelerators, the specialised chips that train and run AI models, for an anticipated one million PFLOPS of computing capacity. A 200-megawatt phase was expected to be handed over by late October 2026, with further handovers planned into 2027.

Demand appears locked in well ahead of time. Envision has secured offtake agreements running through 2029 to 2030, with customers in areas such as autonomous driving and content generation. The two internet companies the campus is built for are not named in the public section of the report. Power in Inner Mongolia runs roughly $0.055 per kilowatt-hour, and rents sit near $40 per kilowatt per month excluding power, a figure the authors say has roughly halved over the past five years.

The West Texas of China

SemiAnalysis describes Inner Mongolia as China's fastest-growing AI datacentre region and compares the area to West Texas: wide open land, plenty of energy and few neighbours to complain. The region holds around 57 percent of China's usable wind resources, capacity factors in Ulanqab sit at 35 to 40 percent, and 25 gigawatts of datacentre projects have already been announced there.

The comparison ends at the politics. Texas is dealing with its highest ever level of anti-datacentre sentiment, according to SemiAnalysis, while Inner Mongolia faces no such backlash. Galaxy is the flagship of Envision's Mission Gobi initiative, which aims to build five gigawatts of green AI datacentre capacity worldwide by 2030. If Galaxy reaches its three-gigawatt goal, this one campus in Ulanqab would make up the majority of that global target.