Data Centres and the Water Question Nobody Priced In
The AI build-out is measured in megawatts. The quieter constraint is measured in litres — and most regions have not done the maths.
The conversation about artificial intelligence infrastructure has been dominated by electricity: how many gigawatts, which grid, whose nuclear reactor. But a second input is moving from footnote to headline in planning meetings across the world — water. Large data centres use water for cooling, and the amounts are not trivial. A mid-sized facility can consume millions of litres a day at peak, comparable to a small town.
The issue is not that the industry uses water. Every industry does. The issue is where. Some of the fastest-growing data-centre corridors sit in regions already under water stress: parts of the American Southwest, northern India, the Middle East, and sections of northern China. When a new facility applies for permits in a drought-prone county, the question residents ask is no longer abstract.
Operators have answers, and some are genuinely good. Evaporative cooling can be swapped for closed-loop systems that recirculate water. Air cooling works in cooler climates, which is one reason Scandinavia attracts builds. Some operators now commit to being water positive, funding watershed restoration to offset consumption. Microsoft, Google and others publish water usage effectiveness figures alongside their power usage figures.
But the reporting is voluntary, the methodologies differ, and the numbers are usually global averages — which can hide the fact that a facility in a wet region offsets one in a dry one. A litre in Ireland is not a litre in Arizona. Regulators are beginning to notice. Several US states and European jurisdictions now require water impact assessments for large new builds, and a handful of projects have been delayed or relocated over local opposition.
There is also a timing problem. Cooling demand peaks on the hottest days — exactly when water systems are most stressed and when grids are most strained. The two constraints compound each other. A facility designed for the average day may be a very different proposition on the worst day of the year.
For communities weighing a data-centre development, the practical questions are becoming clearer. What cooling technology is proposed, and what is its peak-day draw, not its annual average? Where does the water come from, and what else depends on that source? Is the operator committing to a local, auditable water target, or a global corporate average? And what happens to the commitment if the facility changes hands?
For the industry, the direction of travel is visible. Water disclosure is moving from sustainability reports into permitting law. Designs that treat water as free will age badly. The operators who plan for the driest plausible year, rather than the average one, will face fewer surprises.
What is not yet known is how much of the current build-out pipeline has genuinely priced this in. Site selection is often years ahead of construction, and many announced projects have published no water plans at all. The next few permitting cycles will show whether the industry learned the lesson early or will learn it one county hearing at a time.
Every claim above is sourced to a document, a named person, or a record we hold. Where we could not verify a claim, we say so. Read our standards and corrections policy →
