Every time you ask an AI assistant a question, a data centre somewhere runs the computation — and cooling that hardware consumes water. It is one of the least discussed environmental costs of the AI boom, and it has begun to affect municipal water supplies.
Why data centres need water at all
Servers generate heat, and heat has to go somewhere. Many large facilities use evaporative cooling: water is evaporated to carry heat away, much like sweat cooling skin. It is far more energy-efficient than air conditioning, particularly in hot climates, which is exactly why operators favour it.
The catch is in the word evaporative. Up to 85% of the water a data centre draws evaporates and never returns to the supply. This is consumption, not borrowing.
The scale
US data centre water consumption approached one trillion litres a year by 2025, driven largely by AI expansion. Projections for AI servers in the US alone run to 200–300 billion gallons annually between 2024 and 2030.
The local picture is sharper than the national one. Data centres in Northern Virginia used close to 2 billion gallons in 2023 — a 63% rise since 2019. Loudoun County alone, home to roughly 200 facilities, accounted for around 900 million gallons.
The part that affects drinking water directly
Two things matter to households.
First, much of this is potable water. Data centres frequently draw from municipal supply — the same treated drinking water that comes out of your tap. Every litre evaporated for cooling is a litre not available to the community.
Second, what returns is not what went in. Water discharged back into the system carries higher concentrations of dissolved solids: calcium, chloride, silica. Higher dissolved solids affect taste, can reduce crop yields, and at sufficient concentration are toxic to aquatic life. Warmer discharge temperatures add further stress to receiving waters.
If your tap water has begun tasting more mineral than it used to, dissolved solids are the usual explanation — though causes are many and a data centre is unlikely to be yours.
The case for the other side
It would be dishonest to present this as purely negative. AI is being applied to water problems in ways that save considerable volumes: leak detection in municipal networks, where distribution losses commonly run to a fifth of treated water; predictive maintenance that catches main failures before they burst; optimisation of treatment chemical dosing; and modelling that improves how reservoirs are managed through drought.
Some operators have also moved to closed-loop cooling, air cooling in cold climates, or reclaimed non-potable water. These approaches use markedly less drinking water — they are simply more expensive, and adoption is uneven.
Whether the savings outweigh the consumption is genuinely unsettled, and depends heavily on where a facility sits and how it is cooled. A data centre in a water-stressed region is a different proposition from one in a cold, water-rich one.
What it means for a Canadian household
Canada holds a large share of the world’s fresh water and hosts far fewer data centres than the US, so the immediate pressure on Canadian municipal supply is limited. That is changing as capacity is built here, and Alberta and Québec have both attracted significant investment.
The practical takeaway is not to filter your water differently because of AI. It is to read the water quality report your municipality publishes each year, watch the dissolved solids figure, and take an interest in local decisions about who gets to draw on your supply and on what terms.
How to read your water quality report →
Sources are linked inline. Figures cited were current when this article was written and may change.