Direct site footprint66 billion L / year
- Metric
- Modeled water consumed on data-center sites
- Scope
- United States; primarily cooling infrastructure
- Period
- 2023
- Status
- modeled
Are data centers draining community water supplies?
Some facilities consume meaningful water; others use little on site. Cooling system, climate, water source, peak-day demand, drought conditions, and the boundary of the number decide what the impact means.
Key facts, with scope and limits.
Indirect electricity footprintNearly 800 billion L / year
- Metric
- Modeled water consumed at electricity generators serving data-center load
- Scope
- United States electricity supply system; distinct from host-city utility demand
- Period
- 2023
- Status
- modeled
Virginia local variation0.2%–21%
- Metric
- Data-center share of utility water after excluding reclaimed water
- Scope
- Six Virginia water utilities reviewed by JLARC
- Period
- 2023
- Status
- observed
Two water footprints in different places
The chart compares two distinct 2023 national footprints. On-site consumption occurs at data centers. Electricity-source consumption occurs across the power system. Adding them would overstate host-city demand.
LBNL modeled estimates, billion liters per year. The electricity-source figure is reported as nearly 800 billion liters and is plotted here as 800 for scale.
| Category | Modeled consumption (billion L/year) |
|---|---|
| On-site | 66 billion L/year |
| Electricity source | 800 billion L/year |
Use the national baseline to read the local record.
National baseline
Water withdrawal, consumption, delivery, and discharge are different quantities. LBNL's national model provides scale. Cooling design and geography create wide variation, and the indirect power-sector footprint occurs elsewhere.
Local case
National totals cannot answer whether a particular watershed or utility is constrained. A local assessment needs a full-build water balance, source by source, including the hottest day and the drought plan.
What the evidence supports.
Every data center consumes an enormous amount of water.
Overbroad
Evaporatively cooled sites can be significant consumers; air-cooled designs can have very low on-site consumption. LBNL modeled more than one million cooling-and-weather cases rather than one universal rate.
Hundreds of billions of liters come from host-city pipes.
Wrong system boundary
The nearly 800-billion-liter estimate is water consumed across the electricity supply system. LBNL's modeled direct on-site total was 66 billion liters in 2023.
Ask for these local records.
Without these inputs, a project-specific verdict is incomplete. Treat missing evidence as an open question.
- 01Potable, reclaimed, groundwater, and surface-water shares, with enforceable commitments
- 02Annual delivery, consumptive loss, discharge, and metered history kept as separate metrics
- 03Peak-day and peak-hour makeup demand under design weather
- 04Every campus phase and cooling mode, including commissioning and emergency operation
- 05Drought triggers, curtailment priority, backup mode, and the serving utility's capacity plan
Sources used on this page.
- Official reportGrade B2024 United States Data Center Energy Usage Report
United States; historical estimates through 2023 and scenarios through 2028
Modeled national estimates with limited facility-level public data. Location-based emissions and indirect water do not include facility-specific contracts or behind-the-meter supply. - GuidanceGrade AWater-use terminology
United States water-use accounting
Defines accounting terms; it does not quantify data-center use or determine local scarcity. - GuidanceGrade ACooling Water Efficiency Opportunities for Federal Data Centers
Cooling-water efficiency practices for federal data centers
Federal-facility guidance. Technology savings and feasibility remain dependent on the site's climate, chemistry, and operating design. - AuditGrade BData Centers in Virginia
Virginia, primarily FY2021–FY2023, with selected forecasts
Virginia-specific. Several values are stakeholder estimates or model outputs, and future utility-cost scenarios are explicitly uncertain.