USGS county baseline 1985 to 2015 · company disclosures 2024 and 2025

Data center water disclosure and local supply

Where freshwater withdrawals are concentrated, by county

What this map is, and is not. It shows freshwater withdrawals reported or estimated by the U.S. Geological Survey for seven sectors, in million gallons per day (Mgal/d). It is not data center water use. Two optional layers sit on top: data center building footprint from OpenStreetMap, and site water use that Google and Microsoft have disclosed. Neither layer is a measure of water stress: stress depends on demand from all users relative to available supply, and this map does not model that.
View
Year
N N N

Zoom with the + and − buttons, or Ctrl or Cmd plus scroll. Drag to pan. Choosing a county zooms to it. The north arrow shows true north at the selected county, or at the center of the view; on this map it tilts up to about 20 degrees from straight up, and the Alaska, Hawaii and Puerto Rico insets are not rotated.

County detail

Click a county on the map, pick a row from the table, or use Find a county.

National context, not county data

Rosa et al. (2026) count U.S. data centers by months per year in which a global model shows water demand exceeding supply,. This is exposure, not a measured effect on any water user, and the data release has no locations, so it cannot be mapped by county.

0 months1 to 45 to 89 to 12
Baseline 2001-20102,855 (79%)260 (7%)402 (11%)102 (3%)
+3°C warming scenario2,350 (65%)682 (19%)483 (13%)104 (3%)

Out of 3,619 U.S. data centers in the authors' compilation. Their count differs from the OpenStreetMap atlas used for the footprint layer.

Top 15 counties

Method notes and limits
  • Withdrawal, not consumption. USGS reports water withdrawn from ground and surface sources. It does not report how much is returned or evaporated, so withdrawal intensity here is not comparable to consumption figures reported by data center operators.
  • Freshwater only. Saline withdrawals are excluded. Values are groundwater plus surface water, fresh.
  • Seven sectors, harmonized across years: public supply, domestic self-supplied, irrigation (all irrigation types), thermoelectric power, industrial, mining, and livestock plus aquaculture. Thermoelectric is the sum of fossil, geothermal and nuclear categories in 1985 to 1995 and once-through plus closed-loop in 2000 to 2015. Livestock includes aquaculture from 1990 onward.
  • Commercial withdrawals are excluded. USGS reported commercial use only through 1995, so a consistent series does not exist. Data centers would sit partly in commercial and partly in public supply, so this baseline understates the sectors a data center draws on.
  • Double counting. Public supply deliveries serve homes, businesses and industry that also appear elsewhere in USGS tables; the seven sectors here are withdrawals at the source, not deliveries, so they add without overlap.
  • Thermoelectric dominates many county totals. Once-through power plant cooling can be larger than all other uses in a county combined. Use the checkbox to remove it.
  • Trends are approximate at county scale. Estimation methods and sector definitions changed between USGS compilations. National patterns are reliable; single-county jumps between years can reflect method changes or reporting differences.
  • Gray means no usable data. Counties missing from a year's compilation, or whose boundaries changed (old Alaska census areas, Yellowstone Park MT), are gray. Virgin Islands are not mapped. Independent Virginia cities merged into counties are summed into the county; Dade, Shannon and Wade Hampton are mapped to their renamed counties.
  • Series used. The USGS "Revised" county series, which differs slightly from the original 2015 release. The mapped 2015 national total is 280,436 Mgal/d versus 280,689 in the 2015 release (0.09% lower), mostly from revised irrigation values and unmapped territories.
  • 2020 and later. USGS changed to annual modeled estimates on HUC12 watersheds after 2015, so a 2020 county value is not directly comparable to this series and is not shown here.
  • HUC8 outlines are drawn over counties for reference only. Counties are the unit of measurement; watershed boundaries show where county values sit relative to hydrologic units.
  • Class breaks for volume measures are quantiles of nonzero county values pooled across all years, so colors mean the same thing in every year.

Sources: U.S. Geological Survey, Estimated Use of Water in the United States County-Level Data (1985 to 2015 multi-scale archive, "Revised" county series, DOI 10.5066/P14Q6AVK; 2015 release DOI 10.5066/F7TB15V5). County boundaries: U.S. Census Bureau cartographic boundary file, 2015 (1:500,000). HUC8: USGS Watershed Boundary Dataset (simplified). Modeled data center water: Siddik, Shehabi and Marston 2021, Environmental Research Letters 16, 064017 (data DOI 10.7294/14504913; CC0). Modeled 2015 and 2020 withdrawals: USGS National Water Availability Assessment Data Companion (public supply, crop irrigation and thermoelectric models). Aquifers: USGS Principal Aquifers of the 48 Conterminous United States, Hawaii and Puerto Rico (2003, DOI 10.5066/P9Y2HOUJ) and Lovelace et al. 2020, groundwater withdrawals from principal aquifers by county, 2015 (DOI 10.5066/P9EI0KMR). Water-gap context: Rosa et al. 2026, Environ. Sci. Technol. 60, 27780-27791 (CC BY 4.0). Data center locations: IM3 Open Source Data Center Atlas v2026.11.02 (Mongird et al. 2026, DOI 10.57931/17054299), derived from OpenStreetMap contributors (ODbL); crowd-sourced and incomplete, footprint is polygon area and says nothing about water or power use. Alaska, Hawaii and Puerto Rico are shown as insets at different scales.