The cloud is not a cloud. It is a growing archipelago of vast, windowless buildings full of hot computers, and keeping those computers cool takes water — in some facilities, millions of litres a day. As artificial intelligence drives a data centre building boom, water use has moved from an engineering footnote to a planning dispute, with communities asking a pointed question: why should scarce local water cool servers that serve the whole planet?
The industry’s defence is that it is getting more efficient and that digital services displace thirstier activities. Both claims contain truth. But the geography is awkward: data centres cluster where power, land and connectivity are cheap, which often means places — the American Southwest, parts of Spain, the Netherlands’ polders — where water politics were heated long before the first server arrived.
Our reporting on the AI build-out returns often to this physical underside of the digital economy, because the constraints that will shape it are not only chips and capital, but permits, pipes and rainfall.
Why Servers Drink
Computing converts electricity into heat, and heat must go somewhere. Many large facilities evaporate water to reject it — the same principle as sweat — because evaporation is efficient and cheap. Others use closed-loop or air cooling, trading higher electricity use for lower water demand. The International Energy Agency’s work on data centres and energy tracks the sector’s resource footprint, which is growing as AI training and inference add load around the clock.
Measurement itself is contested. Operators report a metric called water usage effectiveness, but figures vary with climate, season and accounting choices — notably whether the water used generating the facility’s electricity is counted. Campaigners argue the true footprint includes that power-station water; operators prefer the narrower boundary. Until reporting standards settle, local debates will keep arguing past each other with different numbers for the same building.
Communities Push Back
From Arizona to Amsterdam, planning hearings that once waved data centres through now interrogate water plans in detail. Some regions have paused approvals; others require facilities to use reclaimed or non-potable water, or to fund watershed restoration. Our technology pages have followed towns that discovered, late in the process, the scale of a project’s projected draw — a transparency failure that poisons trust even where supplies are adequate.
The smarter operators have changed behaviour: publishing water data in advance, designing for the driest plausible year, and siting new campuses where cooling is cheap in every sense. Heat reuse — piping waste warmth into district heating — turns a liability into a local benefit, and several European projects now compete on it. The lesson generalises: infrastructure this large earns its welcome, or loses it, on the resources neighbours can see.
Technology’s Answers
Engineering is responding. Liquid cooling, which circulates coolant directly to the hottest chips, cuts both water and energy use and is becoming standard for AI hardware. Immersion systems go further. Some facilities now modulate their cooling with the weather and the grid, evaporating most on cool, wet days and least when water is scarcest — a flexibility that regulators increasingly reward.
Investigations by the Guardian’s environment desk keep pressure on the gap between pledges and practice, including ‘water positive’ commitments whose accounting deserves the scrutiny it gets. The direction, though, is set: water is now a board-level metric for the cloud industry, reported alongside power and carbon. The buildings going up today will run for thirty years; the ones designed honestly for their watershed will be the ones still welcome in it. The same discipline is beginning to appear in procurement, as large cloud customers ask suppliers for site-level water figures before signing contracts, turning transparency into a competitive feature rather than merely a compliance burden for operators.
