Static snapshot. This site serves two different things: 1,853 data centres reported by OpenStreetMap contributors nationwide, and 13 sites Helios infers from Arizona parcel records, exported on July 29, 2026. The first is a map of what has been recorded; the second is an argued hypothesis with an evidence chain. This is a point-in-time export, not a live view. See meta.json or the methodology.

United States · OpenStreetMap · Lawrence Berkeley Lab

Where data centres are, and how fast they are arriving

Helios tracks 1,853 data centres across the United States, each at a real coordinate and each traceable to the public record it came from. It plots how that number has grown, county by county, against the electricity and water the country is reported to spend on them.

Data centres mapped
1,853
reported by OSM contributors
Counties holding one
276
across 47 states
Net change, 12 months
+345
appeared minus removed
US data-centre power
192
TWh in 2024, reported by LBNL
Three kinds of number appear on this site, and they are not interchangeable. A location is Reported: a contributor mapped it. A date is observed, meaning the month OpenStreetMap first recorded the facility. Never the month it was built, because OpenStreetMap carries no construction dates. A megawatt figure is Inferred, a share of a national total published by Lawrence Berkeley National Laboratory and divided up by building floor area. Only facilities mapped as buildings get one, never a campus land parcel and never a site still under construction. What each of these means.

Data centres on the map, 2012 to today

Full series and national energy →
08331,666tag not yet in use201520162018201920212022202420252026
Data centres recorded in OpenStreetMap at the end of each month. The hatched stretch is before 2017-01, when telecom=data_center was not yet in common use — the near-zero readings there describe the tag, not the country. This is a count of what has been mapped, and OpenStreetMap carries no construction dates.

Where they concentrate

All 323 regions →
CountyMappedFloor area km²Share MW
Loudoun County, VA2392.773,034
Maricopa County, AZ760.961,050
Santa Clara County, CA750.47512
Bexar County, TX670.47518
Prince William County, VA660.921,005
Grant County, WA410.41450
Taylor County, TX400.36394
Morrow County, OR380.55598

The densest county here holds more mapped data centres than most states. Megawatt figures are inferred shares of a reported national total, not meter readings.

What changed lately

Full feed →
  • 2026-06-19 appeared
    Lumen Minneapolis 3 · Hennepin County, MN
  • 2026-06-18 appeared
    DC Blox Data Center · Jefferson County, AL
  • 2026-06-16 appeared
    DataBank SLC3 · Salt Lake County, UT
  • 2026-06-16 appeared
    NTT · Dallas County, TX
  • 2026-06-16 appeared
    DataBank SLC2 · Salt Lake County, UT
  • 2026-06-16 appeared
    DataBank SLC4 · Salt Lake County, UT

A removal means the element stopped matching the data-centre filter in OpenStreetMap. That is not the same as a demolition, and this site never claims it is.

The national picture these counts sit inside

Every published figure →
Electricity

US data centres consumed 192 TWh in 2024 — 3.3× the 58 TWh they used in 2014. For scale, that is a few per cent of all the electricity consumed in the country.

Water

Cooling them consumed 17.4 billion gallons directly in 2023, before counting the water spent generating the electricity itself.

Outlook

LBNL's reference case puts 2030 at 649 TWh. That is a Predicted figure: a scenario, not a measurement. LBNL publishes it as a range, and so does this site.

Start here

what each part of the site answers

A second, deeper dataset: the Arizona study

Browse the site register →

Everything above is a map of what has already been recorded. It cannot see a project before someone maps it, and in practice nobody maps a data centre until the building is standing. The Arizona study is the opposite experiment. In one valley in Maricopa County, Helios reads parcel transfers, permits, assessor classifications and utility filings, clusters them into candidate sites, and argues a confidence score for each with the evidence chain attached.

It currently tracks 13 candidate sites. Every point of every score links back to the document that produced it, and an operator is never named without a direct filing. The highest-confidence site is AZ-CHANDLER-006, at operational.