SemiAnalysis visits an 800MW-class data-center campus in Inner Mongolia and says its first 120MW phase was delivered within a year at about a fifth of US costs

SemiAnalysis, a research firm that tracks data centers in China and worldwide site by site, published on November 7 a report on its September visit to the Galaxy campus in Ulanqab, Inner Mongolia (SemiAnalysis, 2026-11-07). For anyone comparing where large AI data centers can be built fastest and cheapest, SemiAnalysis puts China's edge at a 120MW phase built within a year for about a fifth of US costs. Only SemiAnalysis has made this comparison so far.
Galaxy is planned for up to 800MW of IT capacity. SemiAnalysis compares it with the Abilene campus in Texas that OpenAI uses, which has about 820MW of IT capacity, putting the two on the same scale. SemiAnalysis writes: "Its first 120MW phase was delivered within a year at a fifth of US costs. All four phases are built-to-suit for two internet giants." The first phase came online in August this year, and the full campus is expected to be finished by the end of 2027; the report doesn't name the two Chinese internet companies. SemiAnalysis calls Inner Mongolia China's West Texas: while Texas sees its highest-ever anti-data-center sentiment, Inner Mongolia is accelerating and "faces no such backlash."
Envision, the campus owner, makes wind turbines and energy systems as its core business. Envision builds its own wind farms and runs dedicated lines to the campus, aiming to supply over 60% of the first two phases' power from those wind farms. SemiAnalysis also writes that it is too early to tell whether this vertical integration will become an advantage in China's data-center market.
Not proven yet: only SemiAnalysis says the cost is a fifth of US levels How we checked →
Next to watch: whether Galaxy delivers its later phases on schedule for completion by the end of 2027, and whether any source besides SemiAnalysis puts a figure on its build cost.
Answer check: In September SemiAnalysis said Google's TPUs deliver up to 50% better performance per dollar than NVIDIA's GPUs, and SecondSource's call was that by November 7 a third party with no ties to Google would rerun that performance-per-dollar test on publicly available software. Missed: as of November 7, no such measurement had turned up, so the figure still rests on SemiAnalysis's own test and the TPU cost advantage has not been independently checked. Card →
Source and checks
This item ran in the brief of Nov 8, 2026, alongside 10 other items. How we checked it, the quote checks and what was still missing that day are on that issue's How we checked page.
Cite this item
SecondSource (2026). "SemiAnalysis visits an 800MW-class data-center campus in Inner Mongolia and says its first 120MW phase was delivered within a year at about a fifth of US costs." https://secondsource.io/en/news/2026-11-08-2/. Accessed access date.