# A Screening Atlas of U.S. Data-Center Water-Stress Exposure

## Abstract

Rapid data-center expansion is creating a need for location-specific screening tools that connect facility capacity, modeled water demand, climate exposure, water stress, and local planning context. This brief paper presents a reproducible first-pass atlas workflow using an April 2026 U.S. data-center inventory with location, stage, company type, and MW capacity fields. The method estimates annual water demand from capacity under explicit load-factor and WUE assumptions, assigns Census county and tract identifiers from site coordinates, and ranks sites by a multi-factor water-stress screening score. In the current inventory, 4589 of 4590 sites were geocoded to county and census tract. The inventory represents 424,113.0 MW future capacity basis and 250,272.4 million gallons/year of modeled baseline water demand. Critical-priority sites account for 16,152.0 MW and 9,531.4 million gallons/year. These results are not measured facility water use; they are a transparent prioritization layer for follow-on watershed, public-water-system, NOAA, USGS, and drought joins.

## 1. Introduction

Data centers are increasingly evaluated as coupled water, energy, climate, and community infrastructure. National electricity demand from data centers has grown rapidly, and DOE/LBNL projects continued growth through 2028. Water impacts are more local than electricity demand because cooling design, climate, water source, drought exposure, and public-water-system capacity vary sharply by site. A national atlas is therefore useful only if it preserves site-level geography and separates measured values from modeled or proxy values.

The purpose of this brief paper is to document the first reproducible atlas layer in the repository. The contribution is not a final accounting of U.S. data-center water use. Instead, it is a transparent screening workflow that identifies where the current inventory suggests the strongest need for deeper geospatial joins and validation.

## 2. Data and Methods

The input data are the repository's processed U.S. data-center inventory. Relevant fields include site name, company type, state, region, stage, latitude, longitude, current MW, planned MW, construction MW, and future capacity basis. The source inventory does not provide measured IT load, measured water use, cooling-system type, facility-level PUE, facility-level WUE, water source, or wastewater discharge.

Annual water demand is modeled using the same baseline assumptions already used in the repository: a 75% load factor and Microsoft FY25 Americas WUE/PUE benchmark values. The model uses the WUE form

`annual water = IT load x hours/year x load factor x WUE`.

The water-validation priority index combines modeled annual demand, county withdrawal burden, NOAA heat exposure, U.S. Drought Monitor exposure, source/provider evidence gaps, and construction share. Scores are grouped into low, medium, high, and critical review tiers. The index prioritizes evidence collection and is not a regulatory or hydrologic impact metric.

The first live geospatial crosswalk uses the U.S. Census geocoder. Site latitude and longitude were sent to the Census coordinate geographies endpoint to assign county FIPS and census tract GEOID. The resulting compact crosswalk is stored in the atlas site table; raw API cache files are excluded from Git.

## 3. Results

The 4,590-site inventory represents 424,113.0 MW of future capacity basis. The modeled baseline water demand is 250,272.4 million gallons/year. Critical-priority sites dominate the inventory, with 122 sites, 16,152.0 MW, and 9,531.4 million gallons/year.

The highest modeled state concentrations are TX (13,396 Mgal/year), VA (9,112 Mgal/year), OH (3,991 Mgal/year), OR (2,736 Mgal/year), GA (2,282 Mgal/year). Arizona ranks first in the current screen because large planned capacity overlaps with high heat, drought, and water-stress proxy scores. Texas and Virginia are also major exposure clusters, but for different reasons: Texas combines heat, drought, grid-growth, and large pipeline capacity, while Virginia is driven by exceptional capacity concentration.

## 4. Figures

![Figure 1: Paper Atlas Site Exposure Map](../outputs/figures/paper_atlas_site_exposure_map.png)
![Figure 2: Paper Atlas Top States Water Demand](../outputs/figures/paper_atlas_top_states_water_demand.png)
![Figure 3: Paper Atlas Priority Distribution](../outputs/figures/paper_atlas_priority_distribution.png)
![Figure 4: Paper Atlas Score Vs Water](../outputs/figures/paper_atlas_score_vs_water.png)

Figure 1 maps the site-level screen. Figure 2 shows modeled annual water demand by state. Figure 3 shows how future capacity is distributed across water-stress priority classes. Figure 4 compares modeled water demand against the multi-factor atlas score.

## 5. Discussion

The central result is that capacity concentration matters as much as average facility efficiency. Fleet-level WUE can be low while local exposure remains high if new MW is concentrated in hot, dry, water-stressed, or fast-growing regions. The atlas framing therefore helps move the question from "How efficient is the fleet?" to "Where do planned facilities create the greatest local water-planning burden?"

The current workflow is also useful for stakeholder triage. Data-center operators can identify sites where water-source disclosure, non-potable water, hybrid cooling, or drought contingency planning should be prioritized. Local governments and water utilities can use the planned/construction MW fields to evaluate future demand rather than only present operations. Researchers can use the Census county and tract crosswalk to add ACS, public-water-system, watershed, and drought layers.

## 6. Limitations

This analysis is a screening study. It does not measure facility water use, withdrawals, consumptive use, cooling technology, or source water. The water-stress and climate variables are still proxy fields inherited from earlier repository studies. The score should not be interpreted as a regulatory risk metric or as proof of actual water impact. The next publication-grade version should replace proxies with site-specific joins to NOAA Climate Normals, U.S. Drought Monitor polygons, USGS county water-use data, HUC watersheds, public-water-system boundaries, and ACS/EJ variables.

## 7. Conclusions

The current atlas seed demonstrates that the dataset can support a publishable screening workflow after stronger external joins are added. The immediate scientific value is prioritization: identifying which states, sites, and company/stage combinations deserve deeper hydrologic, climate, and public-water-system analysis. The most promising next paper is a national data-center water-stress exposure atlas that reports scenario-modeled water demand, uncertainty bands, and validated geospatial joins rather than claiming measured water accounting.

## References

- DOE/LBNL. 2024 United States Data Center Energy Usage Report. https://eta.lbl.gov/publications/2024-lbnl-data-center-energy-usage-report
- DOE. Report release on increasing electricity demand from data centers. https://www.energy.gov/articles/doe-releases-new-report-evaluating-increase-electricity-demand-data-centers
- NOAA/NCEI. U.S. Climate Normals. https://www.ncei.noaa.gov/products/land-based-station/us-climate-normals
- USGS. Estimated use of water in the United States county-level data for 2015. https://www.usgs.gov/data/estimated-use-water-united-states-county-level-data-2015
- DOE FEMP. Cooling Water Efficiency Opportunities for Federal Data Centers. https://www.energy.gov/cmei/femp/cooling-water-efficiency-opportunities-federal-data-centers
- EPA WaterSense. Water and Data Centers. https://www.epa.gov/watersense/water-and-data-centers