What Is Water Usage Effectiveness (WUE)?

Water Usage Effectiveness, or WUE, measures how much water a data center consumes for each unit of energy its computing equipment uses. It is expressed in liters per kilowatt-hour, and it captures a side of AI resource use that gets far less attention than electricity: the water needed to keep servers cool and, indirectly, to generate the power they run on.

If PUE tells you how efficiently a facility uses energy, WUE tells you how efficiently it uses water. Both matter for AI, because large models run on hardware that produces a lot of heat, and removing that heat often means evaporating water. Understanding water usage effectiveness helps explain why a text reply or an image has a hidden water cost at all.

Defining WUE

WUE is calculated as the water a data center consumes divided by the energy delivered to its IT equipment, giving a figure in liters per kWh. A facility with a WUE of 1.8 uses 1.8 liters of water for every kilowatt-hour that reaches its servers. A lower number means the site cools its equipment with less water, whether through design, climate, or alternative cooling methods.

The metric was introduced by The Green Grid, the same industry body behind PUE, to give operators a consistent way to report water use. It focuses on water consumed on site, mainly through evaporative cooling, though a fuller accounting also considers the water used upstream to generate electricity.

Why data centers use water

Servers convert almost all the electricity they draw into heat, and that heat has to go somewhere. Many large facilities use evaporative cooling, where water is evaporated to carry heat away, much like sweating cools the human body. It is energy-efficient, which helps PUE, but it consumes water directly, which raises WUE. There is often a trade-off between the two.

On top of on-site cooling, generating the electricity itself uses water. Thermal power plants use water for steam and cooling, so even a data center that used no water directly would still have a water footprint through its power supply. This is why total water estimates for AI, explored in how much water data centers use, combine both sources.

Regional differences

WUE varies enormously by location, more than almost any other efficiency metric. A data center in a cool, humid climate can rely on outside air for much of the year and evaporate little water. One in a hot, dry region may run evaporative cooling hard, driving WUE up, and it does so in exactly the places where water is often scarcest.

Electricity source matters too. A facility drawing on hydropower or thermal generation has a different upstream water profile than one on wind or solar. Because of this, the same AI model can carry a very different water footprint depending on which region serves your request, a point we return to in the environmental impact of AI.

• Cool, humid regions: less evaporative cooling, lower WUE
• Hot, dry regions: heavy cooling demand, higher WUE
• Grid mix: thermal power adds upstream water, renewables add little

Tying it to 3.4 liters per kWh

A useful working estimate for the total water tied to AI is around 3.4 liters per kWh, combining on-site cooling with the water used to generate the electricity. That figure lets you translate energy into water the same way a grid factor translates energy into carbon. It is an approximation that shifts with region and season, not a fixed constant.

Applied to a real request, the math is straightforward. Take the energy a task consumes, including the PUE overhead, and multiply by roughly 3.4 to estimate liters of water. Our carbon footprint calculator uses this approach so water sits alongside energy and carbon rather than being ignored, and you can read the fuller picture in how much water ChatGPT uses.

Lowering the water footprint of AI

Operators reduce WUE through cooling technology and siting. Closed-loop and air-based cooling systems use little or no evaporative water, and placing facilities in cooler climates cuts the need for it. Shifting to renewable electricity also trims the upstream water tied to power generation, addressing both halves of the total.

As a user you do not control cooling design, but you can favor providers that measure and disclose water alongside carbon. A service that reports water per request, and offsets its overall footprint, treats water as a first-class resource rather than an afterthought. That transparency is part of what defines genuinely eco-friendly AI.