Data centers use water through two different pathways: water consumed at the facility for cooling and water consumed indirectly when power plants generate the electricity that runs the servers. Those quantities have different boundaries and should never be presented as one undifferentiated number.
Cooling design, server efficiency, utilization, climate, grid mix and operating practices determine the result. A national estimate can describe the scale of the sector, but it cannot by itself show whether a particular data center is affecting a nearby community’s supply.
As an Amazon Associate I earn from qualifying purchases.
What “data center water use” actually measures
Direct, or site, water consumption
Direct water is consumed at the facility. The main pathway in many water-cooled designs is evaporation from a cooling tower. Some water is also discharged as blowdown so dissolved minerals do not build up in the recirculating loop. Water used for other site functions can be included in a facility’s accounting, but cooling is usually the dominant operational issue discussed in data-center water studies.
The 2024 Lawrence Berkeley National Laboratory (LBNL) report expresses site water-use effectiveness, or WUE (site), as liters of water consumed per kilowatt-hour of electricity used by IT equipment. This is an intensity metric, not a universal amount for every server, workload or building.
Withdrawal is not the same as consumption
A withdrawal is water taken from a river, reservoir, aquifer or utility system. Some withdrawn water may be treated and returned. Consumption is the portion not returned to the immediate water cycle, often because it evaporates or is otherwise unavailable at that location and time. Treating withdrawal and consumption as synonyms can make a facility appear to use more or less water than it actually removes from local availability.
Indirect, or source, water consumption
Electricity generation can consume water even when the data center itself uses little or no water on site. The amount depends on the power plants and grid supplying the facility. LBNL labels this electricity-related metric WUE (source). It is separate from WUE (site), and the two figures should be reported with their accounting boundaries and time periods rather than simply added together.
How much water do U.S. data centers use?
The most specific nationwide water estimate in the available evidence comes from the 2024 United States Data Center Energy Usage Report from LBNL. It is a modeled estimate based on facility types, locations, cooling systems and thermodynamic simulations—not a roll-up of meters from every building.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstall| Measure | Reported value | What the number means |
|---|---|---|
| Direct U.S. data-center consumption, 2023 | 66 billion liters | Modeled site consumption for all U.S. data centers in 2023 |
| Hyperscale and colocation share, 2023 | 84% | Estimated share of the 66-billion-liter total attributed to those facility types |
| Hyperscale direct consumption, 2028 | 60–124 billion liters | Scenario range for hyperscale facilities, not a metered total or a forecast for one named site |
| Indirect electricity-related consumption, 2023 | 4.52 liters per kWh | LBNL’s national average for water consumed in electricity generation serving data-center electricity use |
The 2028 hyperscale range is broad because outcomes depend on assumptions about deployment, power demand, locations and cooling technologies. It should not be read as a guaranteed increase or as a prediction for a particular operator.
LBNL’s electricity-related estimate uses grid and generation data at balancing-authority scale and assumes the local grid mix. It does not incorporate an individual facility’s power-purchase agreement or behind-the-meter generation. Consequently, a facility’s actual indirect water intensity can differ from the national average.
A newer LBNL record, the United States Data Center Energy Usage Report: 2025 Update, is dated June 2026 and updates electricity projections. Its surfaced abstract does not provide revised water totals, so it should not be cited as a new 2026 water forecast.
Why cooling towers consume water
In a water-cooled chiller system, heat from the data hall is transferred to a condenser-water loop. A cooling tower rejects that heat by allowing part of the warm water to evaporate into the atmosphere. More water is discharged as blowdown to control the concentration of dissolved minerals left behind by evaporation.
“A cooling tower system by necessity uses an extensive amount of water because the warm water from the chiller’s condenser water loop is cooled by evaporating water into the surrounding atmosphere (see Figure 1).” — U.S. Department of Energy, Federal Energy Management Program
Evaporation is therefore not a leak or an accidental loss; it is the operating principle that makes the tower effective. Blowdown is a separate, deliberate discharge. The quantity of each varies with heat load, weather, tower controls, water chemistry and the number of times water can be recirculated before treatment is required.
Cooling configurations and their trade-offs
LBNL modeled nine cooling-system categories. The principal configurations described in its report and in Department of Energy guidance have different water, energy and operating profiles.
| Configuration | Direct site-water profile | Important trade-offs and constraints |
|---|---|---|
| Air-cooled chillers | Can minimize on-site water consumption | May use more electricity or require different equipment sizing; performance depends on climate and heat-rejection conditions |
| Water-cooled chillers with cooling towers | Consumes water through evaporation and blowdown | Offers high efficiency and capacity, but requires tower water, treatment and discharge management |
| Direct expansion | Water use depends on the heat-rejection arrangement | Equipment choice, climate and load determine the actual profile |
| Air-side economizer | Can reduce mechanical cooling and associated water use when suitable outdoor air is available | Limited by outdoor temperature, humidity, air quality, filtration and security requirements |
| Water-side economizer | Can reduce compressor operation; water use depends on the connected heat-rejection system | Requires compatible loop design and suitable outdoor conditions |
| Dry or adiabatic cooling | Dry operation uses little direct water; adiabatic assistance adds water during hot periods | Water demand is intermittent and climate-dependent; dry operation can carry an energy or capacity penalty |
| Liquid cooling | Not inherently one fixed water number; the facility’s heat-rejection equipment and whether the coolant is recirculated matter | Can support high-density computing, but plumbing, maintenance, reliability and heat-rejection choices remain site-specific |
There is no universal winner. A design that minimizes direct water may require more electricity, while a highly efficient water-cooled design may be unsuitable where water is scarce or discharge is restricted. Climate, air quality, energy prices, available utility capacity, reliability requirements and site layout all affect the choice.
Free tools Windows power users keep installed
One-click scans. No signup required.
Does AI use a lot of water?
AI workloads can increase resource demand when they require more computation, but “AI” does not have a fixed water footprint. Water per task changes with the server’s efficiency, how fully the hardware is utilized, the cooling system, the climate, the facility’s infrastructure and the electricity source.
A 2025 review by Nuoa Lei, Jun Lu, Arman Shehabi and Eric R. Masanet found that modeled workload-level water use varied by more than 10,000-fold. The review attributed that spread to more than 1,000-fold differences in water consumption per kilowatt-hour of server electricity and roughly 10-fold differences in workload efficiency. It identified server efficiency, grid water-consumption factors, utilization, cooling type, infrastructure efficiency, climate zone, the share of inactive servers and hardware refresh cycles among the key determinants.
That result makes a universal “water per AI prompt” claim misleading unless it states the model, hardware, prompt length, batching, utilization, cooling configuration, location, season and electricity mix. A short inference on an efficient, well-utilized system can have a very different footprint from a large training run on older hardware in a water-intensive climate. The available studies support comparing those conditions, not assigning one permanent number to every AI service.
How operators can reduce direct water use
Department of Energy Federal Energy Management Program guidance describes measures for facility operators. They are engineering and controls decisions, not household water-saving tips.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problems- Set temperature and humidity targets within equipment guidance. Raising or widening set points where the hardware manufacturer permits can reduce unnecessary cooling demand. Limits must be checked against server specifications, condensation risk and reliability requirements.
- Contain hot and cold air. Physical separation and containment reduce mixing between supply and return air, allowing cooling equipment to operate more effectively. Poorly sealed aisles can erase the benefit of otherwise efficient equipment.
- Use air-side economizing when conditions allow. Outdoor air can provide cooling during suitable temperature and humidity periods. Operators must account for air quality, filtration, security and the number of hours the local climate actually permits economizing.
- Use water-side economizing where the system is designed for it. A compatible loop and heat-rejection path can reduce compressor operation in favorable weather, but the resulting water profile depends on the connected equipment.
- Manage cooling-tower concentration and blowdown. Water treatment and controls can permit more recirculation before discharge, while staying within corrosion, scaling, biological-control and equipment limits. Excessive concentration can damage the system; excessive blowdown wastes water.
DOE cautions that savings depend on climate, set points and operating hours. A measure that works in a cool, dry location may provide little benefit in a hot or polluted environment, and changing controls without commissioning can create reliability problems.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Can a data center affect a local water supply?
Yes, it can in circumstances where its withdrawals or consumption are large relative to the available local supply, especially during dry seasons or when several users compete for the same source. But a national average cannot establish that a particular facility is straining its community. Local impact requires local measurements and context.
Information needed for a credible local assessment
| Question | Evidence to obtain |
|---|---|
| How much water does the facility use? | Direct withdrawal and consumption, separated by source, month and season; cooling-tower makeup and blowdown records where available |
| Where does the water come from? | Municipal utility, surface water, groundwater, reclaimed water or another source, including permitted quantities and restrictions |
| What supply is available to other users? | Utility capacity, watershed or aquifer conditions, drought status, environmental flows and competing residential, agricultural and industrial demand |
| What returns to the environment? | Discharge volume, temperature, chemistry, treatment, destination and any reuse or recycling arrangement |
| Does electricity add an indirect burden? | The facility’s grid geography and generation mix, plus any relevant on-site generation; a national average is not a substitute for this information |
Seasonality matters. A facility with moderate annual consumption may create a sharper concern if its peak demand coincides with drought or summer utility stress. Conversely, reclaimed water, storage, low-water cooling and a source with substantial spare capacity can change the local assessment. The conclusion must follow the documented source, timing and competing demand rather than the facility’s name or national statistics.
What is Project Oasis?
A 2024 Department of Energy notice describes Project Oasis as a proposed feasibility project to investigate circulating water from abandoned underground mines in Virginia for data-center cooling. The notice supports describing it as an investigation, not as proven technology, broad commercial deployment or a generally available solution. Any such approach would still need testing for water quality, pumping, treatment, heat transfer, environmental protection, permitting and long-term reliability.
Recommended Free Tools
How to evaluate water-use claims
- Check the boundary. Ask whether the number is direct site consumption, withdrawal, indirect electricity-related consumption or a combined figure.
- Check the unit. Liters per kilowatt-hour is an intensity metric; annual liters is a total. Neither alone tells you the impact on a particular watershed.
- Check the date and geography. A 2023 modeled U.S. average is not a 2026 measurement for one building, and a grid-average factor may not match a facility’s actual supply.
- Check whether it is measured or modeled. LBNL’s national figures are estimates produced from assumptions about facilities, cooling and location, not direct meters at every site.
- Check workload assumptions. AI claims should disclose hardware efficiency, utilization, cooling, climate, electricity source and the workload itself.
- Do not infer a universal ranking. Air, water, dry, adiabatic, economizer and liquid systems trade water against energy, capacity, climate suitability and operational constraints.
The most defensible answer to “How much water does a data center use?” is therefore conditional: identify the site boundary, report direct consumption and withdrawals separately, describe the cooling system, add indirect electricity water only with a stated method, and use local supply data before drawing conclusions about community effects.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
