The Hidden Water Cost of the AI Boom
Published by Water Network Research, Official research team of The Water Network in Business
SEE FULL ARTICLE ATTACHED
Introduction
Every time you run an AI query, stream a video, or back up your files to the cloud, water is consumed. Not inside your device — inside the power plants and cooling towers that keep global data infrastructure running. As the artificial intelligence boom drives an unprecedented expansion of data centre capacity, the water sector faces a new and rapidly growing category of demand that most water professionals have barely begun to consider.
The United States alone holds 43% of global installed data centre capacity, growing at nearly 20% per year. In January 2025, an executive order accelerated AI-focused build-out further, while simultaneously offering exemptions to Clean Water Act permits. The collision between digital infrastructure expansion and global groundwater stress is no longer a future scenario — it is happening now.
How data centres consume water
Data centres use water through two distinct pathways, and understanding the distinction is essential to evaluating any operator's environmental claims.
On-site evaporative cooling
Traditional facilities use cooling towers that evaporate water to dissipate heat from server rooms essentially large-scale swamp coolers. This water is consumed outright: it leaves the water cycle as vapour and does not return to local aquifers or surface water bodies. Industry benchmarks measure this as Water Usage Effectiveness (WUE), expressed in litres per kilowatt-hour of IT load. Evaporative cooling typically achieves a WUE of around 1.8 L/kWh — meaning a 100 MW facility consuming one million kWh per day evaporates approximately 1,800 cubic metres of water daily.
Upstream energy water footprint This is the larger and almost entirely hidden portion. Thermal power plants — gas, coal, and nuclear require substantial water for steam generation and condenser cooling. Approximately 80% of a data centre's total water consumption is embedded in the electricity it draws from the grid. The 2024 US Data Centre Energy Usage Report estimated the indirect water footprint of US data centres at roughly 800 billion litres in 2023 alone — a figure that dwarfs on-site consumption and is projected to increase sharply.
"Amazon's claim that its facilities use 'only a small fraction' of a region's water system is technically defensible but misses the point. Adding a massive new consumer to an already stressed system compounds scarcity regardless of relative share
Tags
Category: Business
- Water Scarcity
- Water Footprint
- Water Cycle
- IT
- artificial intelligence
- Domestic Water Use
- United States