Counting Data Centers: The OpenStreetMap Inventory Behind a National County-Level Comparison
A September 2026 trade association report sorts more than 3,200 U.S. counties into four groups by how many data centers each contains, and runs its comparisons of housing, employment and electricity outcomes across those groups. This note examines the inventory that produces the grouping: the IM3 Open-Source Data Center Atlas, an OpenStreetMap-derived facility file published by Pacific Northwest National Laboratory. The report's counts reproduce exactly from its cited source, in four separate checks. Its arithmetic is correct, its source is public, named and openly licensed, and it states its own limitations repeatedly. The findings here are narrower and prior: what a county-level data center count measures, and how much weight a comparison across such counts can carry. The unit counted is an object tagged in OpenStreetMap, and 1,237 of 1,474 records are individual buildings rather than facilities or campuses, so whether one data center contributes one record or twenty depends on how a volunteer contributor drew it. Between the two Atlas versions the report cites, eight months apart, 55 of 252 counties fall into a different one of the report's four groups. The February 2026 version records no data centers in 58 of 163 counties where EPA held air-program records under NAICS 518210, nor in nine Virginia localities where the state had issued air permits to named data center facilities. The note also audits its own comparison instrument, which earlier drafts did not. Ten of the 79 facilities behind the national coverage count are affirmatively not data centers, including flare-gas crypto mining rigs on oil and gas well pads, a coal-fired power station and a parking lot rehabilitation project. That finding moved the headline figure and withdrew an example an earlier draft had named. Corrections found across five drafts and four independent review rounds are kept in the running text rather than made silently. Eight Python scripts reproduce every figure, standard library only, and every input file is included, so the archive runs cold with nothing to download. The two IM3 Atlas CSV files travel under the Open Database License rather than this record's CC BY; section 10 of the note and the archive README state the per-file terms, with the exact versions and their md5 hashes.
Authors
- Tommy Turner (ORCID: https://orcid.org/0009-0005-7422-9881)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-09-17
- DOI
- https://doi.org/10.5281/zenodo.22819654
- Primary Topic
- Data Analysis with R
- Type
- article
- Field-Weighted Citation Impact
- 0.00