The clearest case: a big uncontrolled turbine fleet in clean air, running.
Campus page →Many AI campuses plan their own gas power. We checked which plants burn gas today, using Sentinel-5P satellite readings of nitrogen dioxide and state and federal plant records.
Most on-site gas is still on paper: only 5 plants burn gas today. Of 42.6 GW of on-site generation filed (permitted or applied for) at 28 campuses that have reached site work or beyond, 2.2 GW is running.
On-site generation filed (permitted or applied for), in GW, at campuses with site work started or further along, grouped by what plant records show. Campuses with no filing show a dash.
Nitrogen dioxide over the Southaven turbine yard, net of the rural background, is about 4.4× last summer’s level. Gas booked on the pipeline implies roughly 760–1,030 MW of IT power draw (scheduled, not metered). And Mississippi’s environmental regulator (MDEQ) wrote in a July 2026 agreed order that the temporary turbines operate at the site.
Texas Gas Transmission's public postings show how much gas is booked each day for delivery to MZX Tech, the xAI company that runs the Southaven turbines. The record opens on July 23, 2026 at 180,000 dekatherms a day. Bookings averaged about 223,400 a day in September: enough gas for roughly 810–980 MW of turbine output, or about 600–820 MW of IT power draw once cooling and losses are taken out. Oct 1–13: xAI's operator booked zero gas on the pipeline for 13 straight days; that is in the pipeline's own daily postings, not missing data. Bookings restarted Oct 14 at a new meter. Satellite NO2 over the site did not drop in those days, so the turbines may have run on fuel from another source, or the NO2 came from other plants in the area (TVA's Southaven plant sits in the same box). The public records can't settle which. From Oct. 15 to Oct. 22 they averaged about 281,600 a day, or 760–1,030 MW of IT power draw, with a high of 307,709 on Oct. 22.
Since Oct. 14, bookings have been above MZX's firm pipeline contracts (193,200 a day as of Oct 1, 2026), so the extra rides on interruptible service or newer contracts; the postings do not say which. These are booked (scheduled) quantities, not metered flow: the pipeline does not publish what was actually burned. The conversion assumes a heat rate of 9,500–11,500 Btu per kWh and a PUE of 1.2–1.35.
Each chart shows NO₂ over the campus (orange) and over open country nearby (blue), January 2025 to October 2026: weekly readings, then daily for the last 90 days, with cloudy gaps filled by straight-line interpolation and lines smoothed (3-week / 7-day averages). The shaded bands are July–September of each year, the season compared. Charts use their own scales. Sorted by the strength of the rise; the tag says what plant records show.
The clearest case: a big uncontrolled turbine fleet in clean air, running.
Campus page →A rise, but likely construction or regional sources, no plant built yet.
Campus page →A rise, but the box overlaps the Colossus 2 turbine yard, so it is probably Colossus 2.
Campus page →A rise, but likely construction or regional sources, no plant built yet.
Campus page →A rise, but likely construction or regional sources, no plant built yet.
Campus page →A rise, but likely construction or regional sources, no plant built yet.
Campus page →A rise from a low base where engines are running; small, so read it as suggestive.
Campus page →A rise, with no plant-record check yet; cause unknown.
Campus page →No clear change; no plant-record check yet.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →No clear change; no plant-record check yet.
Campus page →The plant runs, but the site is in suburban Columbus where traffic and industry already raise NO₂, so the satellite cannot see it.
Campus page →A control: a retired coal plant, no combustion expected, and no change seen.
Campus page →No rise; plant records show engines installed but no sign they burn gas.
Campus page →No clear change; no plant-record check yet.
Campus page →No clear change; no plant-record check yet.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →No clear change; no plant-record check yet.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →No clear rise, and no plant built yet. Consistent.
Campus page →A control: no plant, no change expected, and none seen.
Campus page →The plant runs, but the site sits in the San Antonio city plume, so the satellite cannot pick it out.
Campus page →The latest public record for every campus with on-site gas plans that we checked, from permits, orders, federal generator inventories and court filings. Sorted by planned size within each group.
Contains modified Copernicus Sentinel data 2025–2026.