A low WUE can still mean a lot of water
A data center reports a Water Usage Effectiveness of 0.4 L/kWh.
That sounds efficient.
Now add one missing fact: its IT equipment consumes 100 GWh of electricity each year. At that scale, the ratio represents 40 million liters annually.
A much smaller data center reports a WUE of 1.0 L/kWh. Its IT equipment uses 2 GWh a year, giving it a total of 2 million liters.
The first data center has a WUE that is 60% lower.
It also uses 20 times as much water.
Nothing has gone wrong with the calculation. Water Usage Effectiveness measures intensity, not total demand. It tells us how much water is associated with each unit of energy used by the computing equipment.
That is valuable information.
It is not the same as answering how much water a facility needs, where that water comes from, or what its use means for the place supplying it.
What WUE actually measures
The Green Grid introduced Water Usage Effectiveness in 2011 as a companion to Power Usage Effectiveness. The metric was intended to make water visible alongside the electricity used to support IT equipment.
The current international standard, ISO/IEC 30134-9:2022, defines WUE as a key performance indicator for water consumption during the use phase of a data center. It introduces measurement categories and rules for calculating and reporting the result. ISO lists the 2022 edition as published and still current, while a second edition is under development.
In its simplest form:
WUE = annual water quantity ÷ annual IT equipment energy
The result is usually expressed in liters per kilowatt-hour. Cubic meters per megawatt-hour gives the same numerical value because both parts of the ratio scale by 1,000.
A WUE of 0.5 L/kWh means the facility uses half a liter of water, under its stated reporting definition, for every kilowatt-hour consumed by servers, storage, and networking equipment.
Lower generally means less water per unit of IT energy.
The phrase under its stated reporting definition is doing important work.
Before comparing two WUE figures, the reader needs to know what each facility counted as water, where it placed the boundary, which measurement category it used, and which operating period the figure covers.
Without that information, the apparent precision can be deceptive.
The numerator depends on what counts as water use
Water terminology is less settled in everyday discussion than it looks.
A facility may withdraw water from a municipal network, river, lake, aquifer, or another source. Some may be returned. Some may be discharged elsewhere. Some may leave the local system through evaporation.
These are different physical events.
The U.S. Geological Survey defines consumptive use as the portion of withdrawn water that is evaporated, incorporated into products, consumed, or otherwise unavailable for immediate reuse in the local system. Withdrawal describes water taken from a source. Consumption describes what does not return in a readily available form.
A cooling tower illustrates the distinction. Water enters and circulates repeatedly. Some evaporates while carrying heat away. Some is discharged as blowdown to prevent minerals becoming too concentrated. Fresh makeup water replaces those losses.
A closed circuit behaves differently. Its working fluid may circulate for long periods without being replaced during every cooling cycle. Initial filling, maintenance, treatment, leakage, and occasional replacement still matter, but the same liter passing through the loop ten times has not become ten new liters of input.
This is why closed-loop cooling needs a carefully stated boundary. Water inside a system is not proof of continuing consumption, while a closed loop in the data hall does not reveal whether another stage uses evaporation.
European reporting rules make the boundary explicit. Commission Delegated Regulation (EU) 2024/1364 requires covered operators to report water entering the data-center boundary and used for functions including cooling, power, security, and IT. It also records potable-water input separately.
That method answers a useful question: how much water crosses the facility boundary relative to IT energy?
It does not automatically reveal how much was potable, freshwater, reclaimed water, seawater, discharged, or consumed through evaporation. Those distinctions require additional information.
The denominator can improve while water demand rises
WUE divides water by the energy consumed by IT equipment. It does not divide water by servers, customers, transactions, models trained, prompts answered, or useful computing work.
Consider a facility using 10 million liters of water while its IT equipment consumes 20 GWh.
Its WUE is 0.5 L/kWh.
A year later, the facility expands. IT energy rises to 40 GWh and water input increases to 12 million liters.
Its WUE is now 0.3 L/kWh.
The ratio has improved by 40%.
Total water demand has risen by 20%.
The facility is using less water for every kilowatt-hour reaching the IT equipment. That may be a genuine operational improvement. A local water utility still has to supply another 2 million liters.
The denominator can also improve when a site becomes more heavily utilized. Pumps, controls, and cooling systems may already be running before every available server is busy. As more IT energy passes through the site, fixed or slowly changing water demand is divided by a larger denominator.
WUE can therefore be valuable for tracking the same facility over time. It is less useful as a standalone ranking between facilities of different sizes, functions, climates, and loads.
The denominator also assumes energy reaching IT equipment represents useful activity. An idle server still consumes IT energy. An old processor completing little work per watt still contributes. A facility can report a good WUE while using its computing equipment poorly.
That wider problem sits at the center of our article on data-center efficiency metrics. PUE and WUE normalize supporting resources against IT energy. Neither measures useful work.
The limitation becomes particularly important when WUE is used to estimate the water required by an AI request. Model size, response length, batching, utilization, idle capacity, hardware, and system boundaries all change the allocation. That is why per-prompt water estimates vary so widely.
WUE has no map and no calendar
Two data centers can report the same WUE and create different local consequences.
One may use reclaimed water in a region with abundant supply. The other may draw freshwater from a stressed basin during summer.
The ratio can be identical because WUE measures intensity, not the condition of the water system supplying it.
Location is missing from the formula.
The Green Grid’s Water Usage Impact metric was developed to address this weakness by combining water consumption with local water stress. It does not create a complete environmental score, but it recognizes that the significance of a liter depends partly on competition around it.
Time is missing too.
Annual WUE smooths twelve months into one ratio. A facility may look reasonable across the year while placing its greatest demand on the water system during the hottest and driest weeks, when cooling needs rise and other users are under pressure.
The European Environment Agency measures water scarcity by season and river-basin subunit because annual national averages can hide short but severe periods of stress.
For an operator, an annual value remains useful for consistent reporting.
For a municipality or water supplier, peak daily demand, summer consumption, drought behavior, and the facility’s response to restrictions may matter just as much.
A credible disclosure therefore needs the annual ratio and the operating pattern behind it.
The electricity supply sits outside most site WUE figures
WUE usually focuses on water used within the defined data-center boundary during operation.
Electricity creates another water footprint outside that boundary. Some power stations withdraw or consume water for cooling and other processes. The amount varies by technology, region, and operating conditions.
A data center with little water input at its own site may therefore be associated with water use through its electricity supply.
Lawrence Berkeley National Laboratory found the water intensity of electricity was one of the largest determinants of workload-level water use, alongside server efficiency, utilization, cooling, climate, and facility efficiency. Across the conditions studied, water use varied by more than 10,000 times.
A site WUE cannot reveal that upstream difference.
That is not a reason to abandon it. Operators have more direct control over cooling and water systems inside the site boundary than over every generator supplying the grid.
The problem begins when site WUE is described as the complete water footprint of the data center or the services running inside it. Construction and hardware also sit outside the operating ratio. Semiconductor fabrication, server production, cooling equipment, electrical infrastructure, transport, maintenance, replacement, and end-of-life treatment all require resources. Those effects belong to a lifecycle assessment.
WUE tells us about one stage of one system.
Its usefulness depends on remembering which one.
Seven questions to ask before trusting a WUE figure
1. What is in the numerator?
Does the figure use total water input, freshwater, potable water, net consumption, or another definition? Does it distinguish reclaimed or non-potable sources?
2. Where is the boundary?
Does the number cover the whole data center, one technical space, a mixed-use building, or one cooling circuit? Which measurement category was used?
3. What period does it describe?
Is it an annual operating value, seasonal result, commissioning test, design estimate, or modeled projection?
4. How much IT energy sits in the denominator?
What was the load and utilization during the reporting period? Did the facility expand or change its hardware?
5. What was the total water volume?
A good ratio can accompany a large absolute demand. Annual input, peak daily use, and seasonal consumption belong beside it.
6. What kind of water was used, and where?
Potable water, reclaimed wastewater, groundwater, surface water, and seawater have different infrastructure and local implications.
7. What remains outside the figure?
Does the claim exclude water associated with electricity, hardware manufacturing, construction, maintenance, or other lifecycle stages?
A data center does not need a perfect answer to every question before WUE becomes useful.
It needs enough information for the reader to understand what the ratio supports.
WUE is a deployment metric, not a brand adjective
Policloud develops and deploys physical, modular data-center infrastructure for defined sites.
For that kind of infrastructure, one fleet-wide WUE is less useful than evidence attached to a particular deployment. Cooling configuration, workload density, climate, IT load, water source, reporting period, and system boundary can all change the result.
For an operator, WUE can show whether water intensity improves as equipment, cooling, or operating practices change.
For an infrastructure buyer, it can help compare options using compatible definitions and workloads.
For a local authority or water supplier, it should sit beside total annual volume, seasonal peaks, water source, and drought behavior.
For a customer buying compute, it may form part of a wider resource assessment, but it does not show how efficiently an application uses the hardware.
Each reader is asking the ratio a different question.
No single WUE can answer them all.
Used carefully, WUE reveals how much water supports a unit of IT energy.
Used carelessly, it can make the largest water user on the page look like the smallest.
The difference lies in the information published beside it.
A trustworthy WUE figure carries its total volume, source, location, operating period, and measurement boundary with it.
Only then can the reader reconstruct the water behind the number.
