Data Center Water Usage Statistics: Direct, Indirect, and Local Impacts

Data centers use water onsite and through their electricity supply. The amount varies with climate, cooling design, grid mix, and computing load.
Our cloud computing statistics show how fast that infrastructure is expanding. National totals can appear modest beside agriculture or public supply. Local effects can still be large near stressed watersheds.
US data center water usage reached roughly 65 billion onsite liters during 2023. Lawrence Berkeley National Laboratory produced the estimate. The report projected about 145 to 275 billion liters in 2028. Separate estimates place total US consumption, including broader effects, at 140 to 280 billion liters.
Key Data-center Water Statistics
Diagram
Key Data-center Water Statistics
- Estimated US onsite water use
- Projected US onsite water use
- Projected US total water consumption
- Workload-level variation
- Microsoft global WUE
- Microsoft Americas WUE
- Google 2025 replenishment
| Measure | Figure | Period and boundary |
|---|---|---|
| Estimated US onsite water use | About 65 billion liters | 2023 direct use |
| Projected US onsite water use | About 145 to 275 billion liters | 2028 scenario range |
| Projected US total water consumption | 140 to 280 billion liters | 2028 cited estimate |
| Workload-level variation | More than 10,000-fold | Across server, grid, cooling, and location factors |
| Microsoft global WUE | 0.27 L/kWh | FY2025 owned facilities |
| Microsoft Americas WUE | 0.34 L/kWh | FY2025 owned facilities |
| Google 2025 replenishment | 7.7 billion gallons | Company water projects |
The direct and total estimates use different boundaries. They should not be added.
How Do Data Centers Use Water?
Servers release nearly all consumed electricity as heat. Facilities must move that heat away from processors and other equipment.
Cooling towers can reject heat through evaporation. Other sites use air cooling, liquid loops, or reclaimed water.
Facilities may also use water for humidification, maintenance, and support systems. These uses are smaller than cooling at many large sites.
Electricity generation adds offsite water demand. That portion can exceed onsite cooling under some grid conditions.
What Is the Difference Between Withdrawal and Consumption?
Water withdrawal measures water taken from a source. Water consumption measures water not returned promptly to that source.
A facility may withdraw water, treat it, and return some volume. Evaporated water counts as consumed.
These terms cannot be swapped. A withdrawal figure may be much larger than the consumed amount.
Discharge quality and temperature also affect local systems. Volume alone does not capture every water effect.
What Is Direct Versus Indirect Water Use?
Direct use occurs at the facility. Cooling and humidification account for most direct operational demand.
Indirect use occurs elsewhere. Power stations may consume water while generating electricity for the site.
Hardware manufacturing adds supply-chain water. Semiconductor production uses purified water, but public facility reports often exclude it.
A complete footprint should state which categories it covers.
How Much Water Did US Data Centers Use in 2023?
The LBNL 2024 US Data Center Energy Usage Report estimated about 65 billion onsite liters during 2023.
Its modeled trend rose from roughly 20 billion liters in 2014. Hyperscale and large colocation facilities drove much of the increase.
The same report projected approximately 145 to 275 billion liters in 2028. This range reflects uncertainty about equipment growth and cooling systems.
The report uses a bottom-up model. It combines equipment shipments, energy demand, cooling physics, facility types, geography, and grid factors.
Why Can Local Impact Exceed the National Share?
National percentages spread demand across a large water system. Facilities draw from specific utilities and watersheds.
Data center water usage can therefore create local pressure without dominating national demand.
A new site can add demand where summer supply already faces pressure. Its timing may overlap with residential or agricultural peaks.
Local effects depend on:
- watershed stress;
- cooling technology;
- water source and quality;
- seasonal temperature;
- facility size and utilization;
- electricity-generation mix; and
- drought and infrastructure conditions.
This explains why a small national share can still shape one community's planning.
How Does Cooling Design Change Water Demand?
Evaporative cooling can lower electricity use while consuming water. Air cooling usually cuts onsite water, but can require more power.
Google says water cooling can reduce cooling energy by about 10% compared with air cooling. The best choice varies by climate.
Closed-loop liquid systems can limit continuous water consumption. They still need heat rejection and may use water elsewhere.
Reclaimed wastewater can reduce freshwater demand. Availability and treatment requirements differ by utility.
The cloud infrastructure section covers related choices around capacity, reliability, and operating cost.
What Does Water Usage Effectiveness Mean?
Water usage effectiveness, or WUE, measures annual facility water per IT electricity. Operators usually report liters per kilowatt-hour.
The standard formula is:
Annual water for cooling and humidification ÷ annual IT electricity use
A lower value indicates less onsite water per unit of computing energy. It does not measure the entire water footprint.
The Microsoft datacenter disclosure reported global WUE of 0.27 L/kWh in FY2025. The Americas value was 0.34 L/kWh.
Regional comparisons need context. Temperature, humidity, cooling design, and facility age affect WUE.
Why Should WUE and PUE Be Reviewed Together?
Power usage effectiveness, or PUE, measures total facility energy against IT energy. Lower PUE means less overhead.
A cooling change can improve one metric while worsening another. Air cooling may reduce WUE but raise electricity use.
Teams should also inspect carbon intensity and water stress. A lower national footprint may hide higher local pressure.
No single metric gives a complete facility assessment.
What Are Large Operators Disclosing?
Microsoft reported global WUE falling from 0.30 to 0.27 L/kWh between FY2024 and FY2025.
The company also recorded a 23% year-over-year WUE reduction at Phoenix facilities. It attributes the change to improved temperature and humidity controls.
Google reported replenishing about 7.7 billion gallons during 2025. The company said this represented 78% of annual freshwater consumption.
Replenishment and consumption remain different measures. Projects may restore water in another place or season.
The AI water usage guide explains why model-level estimates need a separate workload boundary.
What Should Communities Ask Before Approving a Facility?
Public review should request specific operating assumptions. Useful questions include:
- What are maximum daily and annual withdrawals?
- How much water will be consumed through evaporation?
- Which source will supply the water?
- Will the facility use potable, reclaimed, or alternative water?
- How will demand change during heat or drought?
- What cooling technology will operate at full load?
- How will actual use be reported after opening?
Answers should include normal and peak conditions. Average use alone can miss infrastructure stress.
Can a Data Center Use No Water?
Some designs use little or no onsite water for routine cooling. Air-cooled and closed-loop systems are common examples.
The full footprint may still contain indirect water. Electricity generation and hardware production can consume water elsewhere.
"Waterless" claims should state whether they cover onsite operations only.
How Can Operators Reduce Freshwater Demand?
Location often produces the largest change. Cooler climates and lower-stress watersheds offer different options.
Operators can also use reclaimed water, improve server utilization, adjust operating temperatures, and upgrade cooling controls.
Workload scheduling may shift computing toward better times or locations. That strategy needs latency and grid constraints.
Disclosure supports better planning. Facility-level withdrawals and consumption should appear separately.
Methodology
This article uses evidence available through 8 November 2026. Primary sources include LBNL, DOE, peer-reviewed studies, and company reports.
We preserve each source's boundary. Direct onsite water, indirect grid water, and supply-chain water remain separate.
Our guide to data-center electricity use covers the energy side of cooling and computing.
FAQs
How Much Water Does a Data Center Use Each Day?
Daily use varies from minimal amounts to millions of gallons. Facility size, utilization, cooling, climate, and water source control the result.
Do Data Centers Use Drinking Water?
Some facilities use potable water. Others use reclaimed wastewater, non-potable sources, air cooling, or closed-loop systems.
Is Water Withdrawal the Same as Consumption?
No. Withdrawal covers water taken from a source. Consumption covers water not returned promptly, often through evaporation.
What Is a Good WUE?
No universal WUE suits every climate and facility. Compare similar sites and review electricity, carbon, and watershed conditions together.
Does AI Increase Data-center Water Use?
AI increases high-density computing and cooling demand. The final water effect depends on location, system design, and electricity supply.
