Data Centers
Loading the map…
Counties
Every county with a recorded action or a mapped site. Select a row to open its record on the map.
Data and sources
Download
- counties.csvOne row per county: action counts, moratorium dates, demographics, vote margin, electricity prices, generating capacity. CC BY 4.0.
- events.csvOne row per community action, with its summary, status, source links and how it was placed in a county. CC BY 4.0.
- codebook.mdEvery column defined, with known caveats.
- locus_zoning_county.csvZoning-text scores per county from LOCUS-v1. Licensed separately under CC BY-NC 4.0, not CC BY; join on the county FIPS code.
- im3_sites_county.csvMapped data center sites per county. Licensed separately under ODbL 1.0, not CC BY.
Join key throughout: the 5-digit county FIPS code. Methods and code are private during pre-publication review.
Sources
| Source | Supplies | License | Retrieved |
|---|---|---|---|
| datacentertracker.org, built by Cam Acosta and George Ingebretsen | Community actions, summaries, source links | CC BY 4.0 | |
| LOCUS-v1 | Zoning text scores | CC BY-NC 4.0 | |
| Census ACS 2023 5-year, gazetteer, county adjacency, place and county-subdivision crosswalks | Population, income, education, county names, land area, neighbors, placing cities and townships in counties | Public domain | |
| Census cartographic boundary file, 1:20m (plotly/datasets mirror) | County shapes on the map | Public domain | |
| County presidential results (tonmcg) | 2024 vote margin | MIT | |
| PUDL (Catalyst Cooperative), EIA Form 861 | County electricity price, 2024 | CC BY 4.0 | |
| EIA Electric Power Monthly, table 5.6.B | State industrial electricity price | Public domain | |
| EIA Form 860 and 860M | Generating capacity (download only) | Public domain | |
| Epoch AI, “AI data centers” | Frontier AI campuses on the map, placed by county and sized by current power | CC BY 4.0 | |
| IM3 Open Source Data Center Atlas v2026.02.09 (PNNL), which contains information from OpenStreetMap, © OpenStreetMap contributors | Mapped data center sites | ODbL 1.0 |
Typeface: Overpass, SIL Open Font License 1.1.
Reading the data
What the map can and cannot tell you.
- An empty county is not a quiet county.
- Actions come from datacentertracker.org, which finds them through news coverage and web research. Places with little local reporting are recorded less often, and coverage before 2025 is thin. The tracker states that AI systems helped synthesize and structure its database and that entries may contain errors.
- Every county moratorium marked “enacted” was audited.
- Every moratorium record placed in a county whose free-text status reads as adopted was checked against its own text and, where needed, its sources; statewide actions are not audited and count for no county. Project approvals and requests that another body act are not counted, and proposals count as pending; each county record shows what the audit found. A moratorium counts for its county whether the county or a city, township, utility district, tribe or other body in it adopted it, and stays counted after it expires. Measures covering only cryptocurrency mining are not counted. Records added after the audit count by their status until audited.
- The latest months are undercounted.
- Actions enter the tracker with a lag, so the most recent bars in the timeline will grow as records arrive.
- Mapped sites are not all sites.
- The circles count data centers that someone has drawn in OpenStreetMap. That favors established facilities: new and under-construction campuses are often missing. A county absent from the site file has no mapped site, which is not the same as no data center.
- Statewide and federal actions are not on the map.
- Statewide actions are listed inside each county record under its state. Federal actions are in events.csv only, except a few that the tracker ties to one place, which count there (the District of Columbia's actions are federal bills). Two local actions could not be placed in a county with confidence and are left out.
- Zoning scores are model-read and cover most counties.
- LOCUS-v1 scores passages of local law with a language model; the county value is the mean over the zoning passages of its county and municipal codes. Hatched counties have no coverage, which is missing data, not a score of zero.
- The county power price is modelled.
- EIA does not report electricity prices by county. The 2024 price divides the revenue of the utilities that serve each county by their sales, both statewide, so neighboring counties under one utility share a price; Texas retail providers report no service territory and are left out. The state price in each record is EIA's figure for the current year to date, a different period.
- Connecticut, Alaska and Puerto Rico are special cases.
- The tracker files Connecticut's actions under its eight legacy counties, replaced in 2022 by nine planning regions; the dataset places each in the region of the town it names (one naming no town stays under its legacy county, in events.csv only). The base map predates the regions, so Connecticut is hatched and reached through search or the table, as are Valdez-Cordova, Alaska and Bedford city, Virginia, whose boundaries changed. Alaska reports elections by district, so its boroughs have no vote margin; nor do Puerto Rico's municipios, which are in the table but off the map. Neighbors follow the Census adjacency file, which counts counties touching only across water.