At a glance
AWS reports a 2025 water usage effectiveness (WUE) of 0.10 litres per kilowatt-hour for its Sydney Region and 0.06 for its Melbourne Region, its first Australia-specific disclosure, against a global AWS figure of 0.12 and an industry average AWS puts at 0.84.
By AWS’s account, its Australian data centres use no water to cool for 95.5% of the year, relying on free-air cooling and switching to evaporative cooling only in peak summer heat.
AWS named its first Australian water-replenishment project: a two-year collaboration with Great Eastern Ranges in Wollondilly Shire, expected to return 32 million litres a year to the Sydney catchment and support 15 threatened species.
AWS says it returned three litres for every four it used in 2025 and is 75% of the way to its goal of being water positive across global data centre operations by 2030.
The numbers land on a sector that uses about 5.5 gigalitres of mains water a year nationally, roughly 0.04% of the total, per CSIRO, where the binding constraint is which catchment the water comes from, not the national volume.
What AWS disclosed
In a 25 June post by Matt O’Rourke, AWS Head of Infrastructure Policy for Australia and New Zealand, AWS reported water usage effectiveness of 0.10 litres per kilowatt-hour for its Sydney Region and 0.06 for its Melbourne Region in 2025. WUE measures litres of water consumed per kilowatt-hour of energy, where a lower number is better. Until now AWS had published a global figure, 0.12 for 2025, but not a breakdown for individual Australian regions.
AWS benchmarks those numbers against a global data centre industry average it cites at 0.84 litres per kilowatt-hour, putting its Melbourne Region at roughly 14 times that average and Sydney at about eight times. By AWS’s account, the Australian results come from free-air cooling that draws outside air directly across the servers for 95.5% of the year, with the system switching to evaporative cooling, which uses and reuses water, only during peak summer temperatures before reverting to air. AWS frames the design with a stadium-parking analogy: infrastructure built for peak events that sits idle most of the time.
Those are operator-reported figures on a methodology AWS sets itself, and the disclosure is the news. No other hyperscaler operating in Australia has published region-level water efficiency for its local facilities. AWS’s earlier Victorian-first recycled water connection at the Melton plant addressed where its cooling water comes from; these figures address how much it uses to begin with.
How the Australian figures compare
Operators have reached low water use by different routes, shaped by cooling design and local climate. The table collects operator-reported figures on different measurement boundaries, so they indicate approach rather than a ranking.
Operator | Reported WUE (L/kWh) | Water approach |
CDC | 0.01 | Closed-loop cooling since 2007; reports about 5 gigalitres a year saved across 16 campuses |
AWS (Melbourne Region) | 0.06 (2025) | Free-air cooling 95.5% of the year; evaporative cooling in peak heat |
AWS (Sydney Region) | 0.10 (2025) | Free-air cooling 95.5% of the year; Melton site to run on recycled water |
AWS (global) | 0.12 (2025) | Targets water positive across data centre operations by 2030 |
AirTrunk | 0.97 (FY24) | 53% of portfolio water from recycled sources; EY limited assurance |
NEXTDC | 2.25 (FY25) | 773 megalitres in FY25; evaporative cooling at some sites, a design that lowers energy use |
Source: Operator sustainability disclosures and the Water Services Association of Australia, FY24-FY25. WUE is litres of water per kilowatt-hour, where lower is better. Boundaries differ between operators, so figures describe approach more than a like-for-like ranking.
The spread reflects design trade-offs. Closed-loop and free-air designs spend little or no operational water but carry their own energy and capital profiles; evaporative cooling spends more water but less energy, which matters in hotter inland sites. CDC reports near-zero operational water from closed-loop cooling, an approach the National AI Plan singled out. AWS’s free-air figures reflect Sydney and Melbourne’s temperate coastal climates, where outside air does the work for most of the year. The same design would draw more water in a hotter location.
Returning more water than it uses
AWS also moved on the give-back side of the ledger. The company named its first Australian water-replenishment project, a two-year collaboration with Great Eastern Ranges to restore degraded bushland and stabilise eroded creek banks in Wollondilly Shire, south of Sydney. AWS says the work is expected to deliver an additional 32 million litres a year to the Sydney catchment by helping the landscape capture, filter and release rainfall, and to support the recovery of 15 threatened species including the glossy black cockatoo, koala and platypus.
The project feeds AWS’s stated goal of being water positive across its data centre operations globally by 2030, returning more water to communities than it consumes. AWS says it reached three litres returned for every four used in 2025, which it puts at 75% of the way to the target. Globally, AWS says it operates 26 data centres on 100% reclaimed water with 130 more contracted, and has announced more than 50 replenishment projects expected to return over 22 billion litres a year. The Australian figures are the first time AWS has attached a local project and a local catchment to that global commitment.
Where the figures sit in the water debate
National water volume is not the constraint. CSIRO, in its submission to the NSW data centre inquiry, put data centre use at about 5.5 gigalitres of mains water a year, roughly 0.04% of the national distributed total and forecast to reach 17 gigalitres by 2030, against about 235 gigalitres for manufacturing. The pressure is concentrated: a handful of fast-growing, drying catchments carry most of the new load, which is why the source of the water and the local network matter more than the aggregate.
AWS’s disclosure is also partial by design. WUE counts water consumed on site; it does not count the water embedded in the electricity a data centre draws, a limit critics of the metric have raised against AWS specifically. The Sydney debate has run on contested numbers, including a widely reported 2025 projection that Reuters noted rested on an undefined “accessible water” figure. Per-region operational figures from a hyperscaler do not settle those questions, but they narrow the part of the argument that runs without data.
What to watch
Three things will show whether this is a turning point or a one-off. First, whether a second hyperscaler operating in Australia publishes region-level water figures; AWS has now set a disclosure bar its peers report against globally but not locally. Second, whether the Great Eastern Ranges restoration delivers the 32 million litres a year AWS projects, a figure that depends on catchment outcomes measured over years, not a meter. Third, whether the supply-side model spreads, after AWS’s Melton recycled-water connection that Data Centres Australia called a template other operators should copy.
AWS paired the water post with its largest Australian renewable-energy commitment to date: nine new power purchase agreements adding 430MW and lifting its national clean-energy capacity to nearly 1GW, by its account. That connects the two sides of the resource question. The embedded-water gap in any WUE figure shrinks as the grid an operator draws on gets cleaner, the same dynamic running through the energy debate, where new data centre load is helping fund the renewables behind falling power bills rather than drawing on existing supply. Our directory tracks AI data centre locations across Australia as operators disclose capacity and their resource commitments.