Method and sources
Built by Dallin. This page says exactly what the map shows, where every number comes from, and what the data cannot tell you.
What the heights mean
Every tower and stack height is the 2D footprint area of tagged data-center buildings, in square feet. It is not power, IT load, floor area or building height. The atlas has no megawatts, no operating status and no floor count, so none are shown or implied.
| Zoom | What you see | Height rule |
|---|---|---|
| National (below about z6) | One hexagonal tower per metro region (Census CSA, else CBSA, else county), stacked by operator | Linear: 1 km of height = 28,571 sq ft. "Compare hubs" switches to square-root heights, scaled so the largest hub stands at 60% of its linear height and smaller hubs stand taller than theirs; the on-screen scale box says so whenever it is on. |
| Middle (z5 to z11.5) | Stacks on H3 resolution 6 hexagons (about 36 km²) | Linear: 1 km of height = 250,000 sq ft |
| Close (z11 and in) | The real building outlines from OpenStreetMap, colored by operator | One flat 12 m for every building. The outline already shows the footprint, so height carries no data. |
Campus polygons are land area. They are drawn flat with dashed edges and never add to a height. Point records, and campuses with no building outline inside them, are drawn as rings. No footprint is invented.
Source
IM3 Open Source Data Center Atlas, v2026.11.02 (PNNL), derived from OpenStreetMap. Mongird, K., Vernon, C., Burleyson, C., Akdemir, K. Z., & Rice, J. (2026). MSD-LIVE Data Repository. doi:10.57931/17054299. License: ODbL 1.0. The features redistributed with this map stay under ODbL.
The file used is the atlas team's own published copy of v2026.11.02 in the IMMM-SFA/datacenter-atlas repository (static/im3_datacenter_footprints.geojson, commit fe4d8420, 2026-11-05). It carries the same buildings, campuses and points as the release, with state, county, operator, name and square feet, but no OpenStreetMap ids. Where a feature is unchanged from v2026.02.09 it keeps that version's OSM id. The other 738 of the 2,001 features (543 buildings, 156 campuses, 39 points) carry stable stand-in ids instead of OpenStreetMap ids; geometry, operator, county and square feet are the published values, and the totals are exact. The totals match the published release: 235.2 million sq ft of building footprint across 124 regions, 49.4 million of it in the Washington-Baltimore region.
Names and grouping
- "Northern Virginia" is used for the Washington-Baltimore CSA only because the data supports it: Virginia holds 98.9% of that region's building footprint (Loudoun County 68%). If that share fell below 75%, the map would use the Census name.
- Eight operator classes follow Kyle Walker's Day 10 grouping. "Ten largest operators" uses this version's top ten by feature count, plus Other and Not tagged.
- Operator names are normalized (for example, Amazon and AWS become Amazon Web Services). Many features have no operator tag in OpenStreetMap and are shown as not tagged.
What is missing
OpenStreetMap is crowd-sourced. Facilities can be absent, mapped only as a point, or mapped as a campus with no buildings. Operators are often untagged. A tall tower means a lot of mapped building footprint, nothing more.
Credits
Map data © OpenStreetMap contributors. Basemap: OpenFreeMap / OpenMapTiles. Rendering: MapLibre GL and PMTiles. Inspired by Kyle Walker's "The data center skyline" (#30DayMapChallenge 2026, Day 10).