How Much Water Do Data Centers Use to Stay Cool?
Every time you stream a video, back up photos or ask an AI chatbot a question, a server somewhere turns electricity into heat. Getting that heat out of the building is one of the biggest jobs a data center has — and a lot of it is done with water. Here is how it works, how much water is involved, and the part of the story that is usually missed.
Why data centers need cooling at all
Almost every watt a server uses ends up as heat. A large data center can draw 100 megawatts or more — as much electricity as tens of thousands of homes — and all of it has to be removed continuously, or the chips overheat within minutes. There are two broad ways to do it:
- Evaporative cooling. Warm water from the building is sprayed through a cooling tower; a small part evaporates and carries the heat away. It is very energy-efficient, but the evaporated water is gone.
- Dry (air) cooling. Heat is pushed into the outside air through radiators and fans, like a car radiator. It uses almost no water, but needs more electricity, especially on hot days.
The physics: how much water evaporation takes
Evaporating one liter of water absorbs about 2,430 kJ of heat at typical temperatures — about 0.67 kWh. So rejecting 1 kWh of server heat purely by evaporation takes about 1.5 liters of water. Cooling towers also have to flush out some water to stop minerals building up (called blowdown), which brings the total to roughly 1.8–2.2 liters per kWh for a site that cools entirely by evaporation.
WUE: the number to look for
The industry measure is water usage effectiveness (WUE), defined by The Green Grid: liters of water consumed on site per kWh of IT equipment energy. Lower is better:
| Type of site | Typical WUE |
|---|---|
| All-evaporative cooling, hot dry climate | ≈ 1.8 L/kWh or more |
| Evaporative, cooler climate or partly air-cooled | ≈ 1 L/kWh |
| U.S. average, 2023 (LBNL estimate) | just over 0.36 L/kWh |
| Air-cooled / dry coolers | near 0 |
The U.S. average is low because many sites use chillers and outside air for much of the year. Lawrence Berkeley National Laboratory expects it to rise to 0.45–0.48 L/kWh after 2023 as large hyperscale and liquid-cooled AI facilities grow.
How much water U.S. data centers use
The most detailed public estimate is the U.S. Department of Energy–funded 2024 United States Data Center Energy Usage Report from Lawrence Berkeley National Laboratory (LBNL):
- U.S. data centers used 176 TWh of electricity in 2023 — 4.4% of all U.S. electricity.
- They directly consumed about 66 billion liters (17.4 billion gallons) of water for cooling in 2023, up from 21.2 billion liters in 2014. That is about as much as 159,000 American families use at home in a year, at the EPA’s figure of 300 gallons per family per day.
- Hyperscale data centers alone are expected to consume 60–124 billion liters a year by 2028.
For a single site, see the data center water calculator: a 100 MW facility at 80% load with all-evaporative cooling (1.8 L/kWh) uses about 3.5 million liters (910,000 gallons) a day.
The bigger, hidden part: water behind the electricity
Power plants use water too — steam plants evaporate it in their own cooling towers, and hydro reservoirs lose it to evaporation. LBNL estimates the indirect water footprint of U.S. data centers’ electricity at nearly 800 billion liters in 2023 — about 12 times the water used on site, or roughly 4.5 liters for every kWh they drew from the grid.
That changes the conversation. A data center that switches to air cooling to save water uses more electricity — and if that electricity comes from water-hungry power plants, the total can go up. Clean wind and solar power, which need almost no water to generate, cut both.
What about AI?
AI training and inference run on dense racks of GPUs that produce far more heat per rack than ordinary servers. Researchers at the University of California, Riverside (Li et al., “Making AI Less Thirsty”) estimated that:
- Training GPT-3 in Microsoft’s U.S. data centers could directly evaporate about 700,000 liters of fresh water — about 5.4 million liters including the water behind the electricity.
- Answering roughly 10–50 medium-length responses consumes about a 500 mL bottle of water, depending on when and where the model runs.
These are model estimates, not meter readings, and newer hardware is more efficient — but they show why water has become part of the debate about AI.
How the industry is reducing water use
- Siting in cooler climates, where outside air can do much of the cooling for most of the year.
- Direct-to-chip liquid cooling in closed loops, which moves heat efficiently and can reject it with dry coolers.
- Using recycled or non-potable water instead of drinking water.
- Reusing waste heat — some European data centers feed district heating networks.
- Thermal storage — chilling water or making ice when conditions are favorable. See could ice cool data centers?
- Buying clean electricity, which also shrinks the much larger indirect water footprint.
Figures are estimates from the sources below. Individual data centers vary widely with climate, cooling design and how busy they are.
Frequently asked questions
How much water does a data center use per day?
It depends on size and cooling. A 100 MW data center at 80% load that cools entirely by evaporation (about 1.8 liters per kWh) uses roughly 3.5 million liters, or about 910,000 gallons, a day. An air-cooled site of the same size uses almost none on site.
How much water did U.S. data centers use in 2023?
Lawrence Berkeley National Laboratory estimates U.S. data centers directly consumed about 66 billion liters (17.4 billion gallons) in 2023, plus nearly 800 billion liters indirectly through the electricity they used.
What is WUE in a data center?
Water usage effectiveness (WUE) is the liters of water consumed on site per kilowatt-hour of IT equipment energy. The U.S. average was just over 0.36 L/kWh in 2023; fully evaporative sites in hot climates can exceed 1.8 L/kWh.
Why do data centers evaporate water?
Evaporation is one of the most energy-efficient ways to remove heat: each liter of water that evaporates carries away about 0.67 kWh of heat. Cooling towers use this to dump server heat into the air.
How much water does ChatGPT use?
Researchers at UC Riverside estimated that a GPT-3-class model consumes about a 500 mL bottle of water for every 10 to 50 medium-length responses, including water used to generate its electricity. It is an estimate that varies with location and time.