Water has become one of the most contested inputs in the data centre debate. Power gets the headlines, but cooling water decides where many sites can actually be built. This guide answers how much water data centres use, where it goes, and why AI workloads are pushing the question to the front of every planning decision.
How much water do data centres use?
Data centre water usage refers to the freshwater a facility draws to cool its servers and, indirectly, the water used to generate the electricity it consumes. A large data centre can use hundreds of thousands of litres a day for cooling alone. Consumption depends on the cooling method, the local climate, and how the facility's power is generated.
That figure splits into two parts. Direct water is what the site draws on-site for cooling. Indirect water is consumed upstream at power stations, since most thermal and hydro generation uses water to produce electricity. Both matter to a community deciding whether to approve a site.
A single hyperscale campus can consume on the order of a few million litres per day at peak, with operators that report their figures averaging roughly two million litres a day per large site. Industry-wide figures vary widely by region and reporting standard, so treat any single number with caution. In the United States alone, data centres are estimated to draw on the order of 1.7 billion litres of water a day across the sector.
Why do data centres need so much water?
Servers generate heat. Push that heat past a threshold and hardware throttles or fails. Cooling is not optional, it is the cost of keeping the building running. Water is one of the cheapest and most effective ways to move that heat out of the building.
The most common method is evaporative cooling. Warm air or water passes through a cooling tower, some water evaporates, and that phase change carries heat away. Evaporation is efficient, but the evaporated water is gone. It does not return to the local system the way water through a closed loop does.
This is why a hot, dry region can be both attractive and risky for an operator. Dry air makes evaporative cooling more effective, yet dry regions are exactly where water is scarce. The engineering logic and the community's water security pull in opposite directions.
What is water usage effectiveness (WUE)?
Water usage effectiveness measures how much water a data centre uses per unit of computing energy. It is expressed in litres of water per kilowatt-hour of IT energy. A lower WUE means a facility uses less water to do the same amount of computing.
Industry-average WUE is often cited at around 1.8 litres per kilowatt-hour, though efficient new builds report far lower, with the leading hyperscalers under half a litre. WUE sits alongside PUE (power usage effectiveness) as one of the two headline efficiency metrics the sector reports. The two often trade off against each other. Cutting water use can raise energy use, because air-based cooling consumes more power than evaporative cooling.
WUE only counts on-site water. It does not capture the water used to generate the facility's electricity. A site can post a low on-site WUE while still carrying a large indirect water footprint through its grid power.
How do AI data centres change water demand?
AI training and inference run at far higher rack densities than traditional workloads. A conventional rack might draw 5 to 10 kW. An AI rack running GPU clusters can draw 50 to 150 kW. More power packed into less space means more concentrated heat, and more cooling demand per square metre.
That density is driving a shift toward liquid cooling, where coolant runs directly to the chip rather than cooling the whole room. Direct-to-chip and immersion cooling can cut total facility water use compared with evaporative tower cooling, because closed loops recirculate rather than evaporate. The trade-off is higher upfront cost and a more complex build.
So the relationship between AI and water is not one-directional. AI raises the heat load per site. The cooling technology AI demands can, when it is closed-loop, lower water consumption per unit of compute. The net effect depends on which method an operator chooses and how the local grid generates power. Researchers have warned that the scale of AI growth could push global AI-related water withdrawal to between 4.2 and 6.6 billion cubic metres a year by 2027, roughly half the United Kingdom's annual freshwater use, which is why cooling design at each site matters so much.
Can data centres reduce their water use?
Yes, and several proven approaches are in use. The choice usually comes down to cost, climate, and local water availability.
Closed-loop and air-cooled systems recirculate coolant instead of evaporating it, cutting on-site freshwater draw sharply. Some operators use recycled or non-potable water, drawing on greywater or treated wastewater rather than drinking-water supplies. Others site facilities in cooler climates where free-air cooling works for much of the year, removing the need for water-intensive chilling.
Each option has a catch. Air cooling uses more electricity, which can raise the indirect water footprint if the grid is water-intensive. Recycled water needs nearby supply infrastructure. Cool climates may sit far from the population centres that need the compute. There is no single answer, only the right answer for a given site.
How does water usage affect where data centres get built?
Water access is now a primary site-selection and planning constraint, alongside power and connectivity. A site with abundant power but no secure water allocation may never reach approval. In water-stressed regions, planning authorities increasingly scrutinise the cooling method and the projected draw before granting consent.
In Australia, water allocation is governed at the state level and tied to catchment conditions that change with drought. A proposed site's water demand can become a live planning issue, particularly in regions already under allocation pressure. The same dynamic plays out in the constrained corridors of the United Kingdom and other mature markets.
For anyone buying, selling, or developing a data centre asset, water status now sits beside grid connection as a fact that decides feasibility. A powered site without a credible water plan carries real planning risk. This is one reason the public record on water allocations and environmental referrals matters as much as the power data.
Frequently asked questions
How much water does a data centre use per day?
A large data centre can use anywhere from hundreds of thousands to several million litres a day for cooling, with reporting hyperscalers averaging around two million litres a day per large site. The figure varies with cooling method, climate, and facility size.
Do all data centres use water for cooling?
No. Facilities using closed-loop liquid cooling or air-based cooling draw far less on-site water, or almost none. Evaporative cooling is the most water-intensive common method, and it remains widespread because it is efficient and cheaper to run.
What is a good WUE for a data centre?
A lower WUE is better. The industry average is around 1.8 litres per kilowatt-hour, while efficient modern builds report well below 1.0 and the best hyperscalers under half a litre. The right target depends on climate and cooling design, and WUE should be read alongside PUE, since reducing one metric can raise the other.
Does AI use more water than normal computing?
AI workloads run at higher rack densities and generate more heat per site, which raises cooling demand. Whether that means more total water depends on the cooling method. Closed-loop liquid cooling, which AI density often requires, can lower water use per unit of compute compared with evaporative cooling.
Is data centre water use regulated in Australia?
Water allocation in Australia is managed at the state level and tied to catchment and drought conditions. A data centre's projected water draw is assessed in planning, and in water-stressed regions it can be a deciding factor in whether a site is approved. In New South Wales, where most of the country's capacity sits, Sydney Water has estimated the sector could reach 15 to 20 percent of the city's supply by 2035, and federal and state authorities are now setting expectations on how operators source cooling water.