How much water does a data center use?
Published 22 September 2026
Anywhere from almost none to several million gallons a day, and the range is not explained by size. Two facilities with identical computing loads can differ by more than an order of magnitude in water use, because the number is set by the cooling design, not by how much computing happens inside. "Data centers use millions of gallons of water" tells you almost nothing on its own.
The number that actually compares facilities
Published data center water usage figures are almost always a total: gallons per day, or millions of gallons per year. That is a measure of how big a site is. The industry metric that controls for that is Water Usage Effectiveness (WUE): annual water consumed, in litres, divided by the IT equipment's energy use in kilowatt-hours.
| WUE (L/kWh) | What it implies |
|---|---|
| ~1.8–1.9 | Around the reported industry average. Evaporative cooling doing most of the work. |
| 0.2–0.5 | Hybrid, or evaporative used only on hot days |
| under 0.05 | Best-in-class. Closed-loop or air-cooled: essentially no ongoing consumption. |
| 0 | Fully closed-loop or air-cooled, filled once |
The gap between average and best-in-class is roughly thirty-fold. No other single design decision in a data center moves a number that far, and it is why a per-facility gallons figure cannot be compared with anything unless the cooling method is attached.
Where the water actually goes
Evaporative and adiabatic cooling: the thirsty one
Water is sprayed or run across media and allowed to evaporate. Evaporation carries heat away extremely efficiently, which is precisely why it is used. The cost is that the water leaves: reported figures commonly put 70–85% of it lost to the atmosphere, never returning to the local supply.
This is the design behind the alarming numbers. A facility on evaporative cooling in a hot, dry climate is genuinely consuming a great deal of water, and the hotter and drier the site, the more it needs.
Closed-loop liquid cooling: thirsty once
The same water circulates indefinitely, carrying heat to a chiller or dry cooler that rejects it to the air. There is an initial fill and periodic top-ups for losses, and that is broadly it. Annual consumption is small enough that WUE approaches zero.
The trade is energy: rejecting heat without evaporation takes more electricity. Which leads to the part that most coverage misses entirely.
Air cooling: no water, more power
Chillers and CRAC units with no evaporative stage use essentially no water and more electricity than any other option. In a cool climate, free cooling for much of the year makes this genuinely efficient. In Phoenix in August it is expensive.
The trade-off nobody mentions: water and power pull in opposite directions. Evaporative cooling is used because it lowers a facility's PUE: it is the energy-efficient choice. Closed-loop and air cooling raise PUE and cut water to near zero.
So a data center optimised hard for energy efficiency is often the one using the most water, and a facility reporting an excellent PUE may be reporting a poor WUE for exactly that reason. A site can be criticised on either number depending on which one somebody chose to publish.
Consumed versus withdrawn
These are different quantities and they are routinely conflated in the same sentence.
- Withdrawn: taken from a source. Some returns, often warmer.
- Consumed: evaporated or otherwise removed from the local supply. It does not come back.
A closed-loop facility may withdraw a large volume once and consume almost nothing thereafter. An evaporative facility consumes most of what it withdraws, continuously. A figure quoted without saying which it is cannot be interpreted, and the larger-sounding number is usually withdrawal.
And whose water is it?
Potable municipal water and reclaimed or non-potable water have very different implications for a local supply, and the same gallons figure can describe either. Several operators have moved sites onto reclaimed water specifically to remove them from competition with drinking supply: the consumption figure barely changes, and the impact does.
Location matters more than volume for the same reason. A million gallons a day beside a large river in a temperate climate and a million gallons a day in a water-stressed desert basin are not comparable facts, even though they are the same number.
So what is a reasonable figure?
With all of that attached, the commonly reported ranges are:
| Facility | Reported daily water use |
|---|---|
| Typical enterprise data center | Hundreds of thousands of gallons: often quoted at around 300,000, roughly a thousand households |
| Large hyperscale site, hot climate, evaporative | 1–5 million gallons |
| Closed-loop or air-cooled site of comparable size | Near zero ongoing |
Treat these as orders of magnitude rather than measurements. Reported data center water consumption comes from surveys and operator disclosures with inconsistent boundaries: some count only cooling, some include the water embedded in the electricity the site consumes, which is a large and separate number. Uptime Institute's survey work has reported that only a minority of water-cooled facilities exceed 16 million gallons a year, which sits well below the figures that circulate most widely.
Put your own load through it: the data center cooling load and water use estimator gives the range for each cooling design at the same size, with the PUE each one costs alongside it.
The questions that make a number meaningful
If you are handed a water figure for a facility, these four turn it into information:
- What cooling design? Evaporative, hybrid, closed-loop or air. This alone explains most of the variance.
- Consumed or withdrawn?
- Potable or reclaimed?
- What is the WUE? It is the only figure that compares one site to another.
The honest answer to "how much water does a data center use" is that it depends on decisions made years before the building opened, and that the facilities using the most water are frequently the ones that were optimised hardest for energy efficiency.
The electricity side of that same trade works the same way, and its published figures go wrong for a closely related reason: how much electricity a data center uses covers why most quoted numbers are design capacity rather than consumption.
For what these ranges mean against real projects rather than in the abstract, every data center in Minnesota is listed with its disclosed capacity converted into a water figure under each cooling design: including the state's 100-million-gallon review threshold, which catches a 30 MW site and misses an 800 MW one.
Sources
- Water usage effectiveness: the metric's definition
- Data Centers and Water Consumption: Environmental and Energy Study Institute
- Data Center Water Use: MOST Policy Initiative science note
- Calculating the impact of water usage on data center costs and sustainability: Vertiv white paper (vendor, but the methodology is explicit)
Figures are ranges from published surveys and operator disclosures, read in September 2026. Where sources disagree, and on this topic they frequently do: the wider range is given rather than the more quotable number.