Sydney Water, the state-owned corporation that supplies drinking water, wastewater, and recycled water to more than 5 million people, has fielded applications from data centre developers wanting connections of up to 40 million litres a day for a single facility.
That is roughly sixteen Olympic swimming pools, every day, for one building.
That's a lot of water.
Data Centre Australia commissioned analysis puts direct on-site water use by Australia's data centres at about 5.5 gigalitres a year.
That's a lot of water, or is it?
This equates to roughly 0.04 per cent of national use, or a seventh of what the country's public swimming pools get through. They account for 0.7 per cent of Sydney's supply and 0.2 per cent of Melbourne's.
Agriculture as an industry takes two-thirds of Australia's water alone.
Sydney Water has told the New South Wales parliamentary inquiry that if proposed facilities proceed with water-intensive cooling, cumulative demand could hit 250 million litres a day by 2035, or 15–20 per cent of the city's supply.
Industry modelling puts Sydney closer to 1.9 per cent by 2030.
So how have we ended up in a space where the water board is saying one thing and the data centre industry is saying something completely different?
Well, I've looked into it, and they're making different assumptions about one variable: how the buildings are cooled.
Where the water actually goes
Nearly all the electricity a data centre draws, and these things draw an enormous load of power (more on that in the future), ends up as heat, and that heat has to go somewhere; otherwise the parts will literally melt.
There are only two destinations for heat that needs to be cooled — water, or air.
Evaporative cooling sends water that’s been heated up by the data centre into air via a cooling tower. Hot water evaporates, and during that phase change, from liquid to air, the heat from the data centre is carried off, and the system is kept cool.
It is thermodynamically brilliant and cheap on electricity, which is why it has been the Australian default. But the evaporated water does not come back. Around 70 to 80 per cent of what enters an evaporative system leaves as vapour; the rest leaves as blowdown, because minerals concentrate in the remaining water and the system must periodically dump some to prevent scaling. An evaporatively cooled facility therefore needs a continuous, one-way, twenty-four-hour flow of new water. Thirsty!
Dry cooling pushes the heat into air instead, through fans and radiator-like coils — similar to your fridge at home. This is a closed-loop system as far as water is concerned. It is filled once, then recirculates indefinitely, with no evaporation, no blowdown and no continuous makeup.
Dry cooling spends electricity to save water. A dry cooling system will perform worst on exactly the hot afternoons when the load peaks, and the extra generation needed means water gets consumed at the power station instead.
Hybrid systems do both. For the most part, it runs dry, and falls back on the evaporative assist only when it needs the extra oomph, which, in Australia, means the water draw peaks on the hottest days of the year.
Direct-to-chip liquid cooling, which AI rack densities are forcing on the industry, runs coolant through a sealed inner loop. That loop is closed, similar to a water-cooled PC. But the heat still leaves the building through an outer facility loop, and that loop will need evaporative, dry cooling, or a combination of both.
What the market is actually doing
Cooling architecture is not a mandatory disclosure in Australia, and the New South Wales consultation paper released in March 2026 concedes that precise water demand data is not available during the planning phase at all.
What we have instead is voluntary Water Usage Effectiveness reporting, which gives us litres per kilowatt-hour of computing.
Australia's largest operators publish it in their sustainability reports, so we can see their numbers.
CDC has run closed-loop cooling since 2007 and reports 0.01 litres per kilowatt-hour. AirTrunk (before being taken private) reported a WUE of 0.97 in FY24, with 53 per cent of its water recycled. NEXTDC reported 2.25 in FY25, on 773 megalitres.
That is a two-hundred-fold spread in water usage effectiveness inside one market, all because of a different choice of cooling process.
Since late 2025, New South Wales authorities have demanded water figures and imposed a power efficiency cap of 1.3. So far, we've had six projects face the new rules.
NEXTDC's S7 at Eastern Creek is a 612-megawatt facility that OpenAI plans to use. Planning documents lodged in April showed it was intended to cool the servers with Sydney's treated sewage water, meaning less strain on the drinking water grid.
It dropped the plan because it could not get planning permission for the pipeline. The fallback is closed-loop liquid cooling, which cuts water use sharply but costs considerably more electricity, drawn from the power network instead, handing the problem to another utility.
Transgrid says it has limited capacity for large new connections.
So Sydney is moving to closed-loop dry systems. In Victoria, where water is cheaper and requirements are currently looser, evaporative and hybrid systems are still argued in state parliament to be the preferred option for hyperscale builds.
But Australia's fleet of roughly 162 operational centres is small and legacy. But it's absolutely ramping up right now. The pipeline matters more than the stock, and within that pipeline, when it comes to water usage, the debate should be focused on how they're cooled.
Because once built, a data centre runs continuously and cannot meaningfully change how it cools.

