A sustainable data centre runs the same compute on less energy, less water and lower-carbon power than a conventional one. It is measured, not claimed. The benchmarks are power usage effectiveness (PUE), water usage effectiveness (WUE), the carbon intensity of its electricity, and what it does with waste heat. Sustainability here is an engineering and reporting outcome, not a logo.
The word "green" gets attached to almost every new site. Underneath the marketing, sustainability comes down to a short list of numbers a buyer or planner can verify. This guide explains those numbers, why power and water now decide where a data centre can be built at all, and how to tell a genuinely efficient site from a rebranded one.
What does sustainable mean for a data centre?
A sustainable data centre minimises three things for each unit of useful compute: energy waste, water consumption and carbon emissions. It is judged on measured efficiency ratios and verified power sourcing rather than intentions.
In practice that means four questions. How much of the incoming power reaches the servers versus the cooling and overhead? How much water does it draw to stay cool? Where does its electricity come from, and how carbon-intensive is that supply? What happens to the heat it produces? Answer those four and you have a sustainability profile that holds up to diligence.
How is data centre efficiency measured? (PUE and WUE)
The two headline metrics are PUE and WUE. Both are ratios, so lower is better, and both are easy to compare across sites.
Power usage effectiveness (PUE) divides total facility energy by the energy delivered to IT equipment. A PUE of 1.0 would mean every watt reaches the servers and nothing is lost to cooling, lighting or power conversion. Real facilities sit above that. The Uptime Institute's annual survey has put the global average at around 1.56 for several years, while purpose-built efficient sites target closer to 1.2 or below, and the best hyperscale fleets now report figures near 1.1.
Water usage effectiveness (WUE) divides annual water consumption by IT energy, usually expressed in litres per kilowatt-hour. Evaporative cooling uses water to shed heat cheaply, which lowers energy use but raises water draw. A low PUE achieved through heavy evaporative cooling can hide a high WUE. The two metrics trade against each other, so read them together.
A site that quotes one number and stays quiet on the other is telling you something. Ask for both, plus the measurement boundary and the period they cover.
Why does water matter so much now?
Cooling is where a data centre meets the local environment, and water is the constraint that increasingly decides planning outcomes in Australia. Large sites can consume substantial volumes through evaporative cooling, often drawing on potable supply unless designed otherwise.
The planning friction is real. Water authorities and councils now scrutinise cooling water demand as part of approval, particularly in drought-exposed regions. In New South Wales, which holds the largest concentration of data centres in Australia, Sydney Water has estimated that data centres could account for 15 to 20 percent of the city's water supply by 2035, and both federal and state authorities have begun setting expectations on how operators source cooling water. A site with no credible water plan can stall regardless of its power position.
Designs that reduce or remove water draw are becoming the default for new build. Air-cooled and closed-loop systems recirculate rather than evaporate. Direct-to-chip liquid cooling moves heat with far less water than open evaporative towers. Some sites use recycled or non-potable water to avoid competing with drinking supply. For buyers, water strategy is now part of underwriting, not an afterthought.
What role does renewable energy play?
Electricity is the largest source of a data centre's carbon footprint, so the carbon intensity of its power supply matters more than almost anything else. A highly efficient facility running on a coal-heavy grid can still be high-emission. A modestly efficient one on clean power can be the opposite.
Operators reach lower-carbon power three main ways. They sign power purchase agreements (PPAs) that fund new renewable generation. They buy renewable certificates to match consumption. Or they site facilities near abundant clean supply. These differ in quality. A PPA that adds new wind or solar to the grid is a stronger claim than certificates bought after the fact.
The harder target is 24/7 carbon-free energy, matching clean supply to demand every hour rather than annually. Google has committed to running on round-the-clock carbon-free energy by 2030, and Microsoft and Amazon have set net-zero and renewable-matching goals on similar horizons. On the Australian grid, where renewable share varies hour to hour, hourly matching is a meaningful test of a sustainability claim rather than a marketing line.
How does cooling design affect sustainability?
Cooling is typically the second-largest energy draw after the servers, so cooling design drives both PUE and WUE. The shift toward AI workloads is forcing change because high-density racks produce more heat than traditional air systems can handle efficiently.
Air cooling moves chilled air across servers and works well at lower rack densities. Above roughly 30 to 40 kW per rack it becomes inefficient and bulky. Liquid cooling, including direct-to-chip and immersion, carries heat away far more effectively at high density. Liquid systems support AI compute and cut the energy spent moving air, with direct-to-chip designs reported to lower cooling energy by roughly 40 to 60 percent against traditional chilled-air systems, and in closed-loop form they sharply reduce water use.
Waste heat reuse closes part of the loop. The heat a data centre rejects can warm nearby buildings, district heating networks or industrial processes. Reuse is common in parts of northern Europe, where district heating is widespread, and earlier-stage in Australia, partly because the heat sinks and district networks that absorb that heat are far less developed here.
How do you spot greenwashing?
Treat sustainability as a measured claim with evidence behind it. Vague language without numbers is the first warning sign. "Powered by renewables" means little without knowing whether that is a new-build PPA or after-the-fact certificates, and whether matching is annual or hourly.
A short diligence checklist runs through five points. Check for published PUE and WUE, with the measurement boundary and reporting period stated. Check the source and quality of renewable claims, separating an additional PPA from certificate matching. Check the water source and cooling type, with potable draw separated out. Check the carbon reporting scope, whether it covers supply-chain and embodied emissions and not only operational power. And check for third-party verification or a recognised efficiency standard rather than self-assessment. In Australia the reference standard is NABERS for Data Centres, which rates infrastructure, IT equipment and whole-facility performance, and which the federal government requires at five stars for facilities hosting agency workloads.
If a site cannot produce these on request, the sustainability claim is a brochure, not a fact.
Frequently asked questions
What is a good PUE for a data centre?
Lower is better, and 1.0 is the theoretical floor. The global average sits around 1.56 and has barely moved in years, older facilities often run well above 1.5, while efficient purpose-built sites target around 1.2 or lower and the best hyperscale fleets reach close to 1.1. Always check the measurement boundary, because a narrow boundary flatters the number.
Do data centres use a lot of water?
They can, mainly through evaporative cooling. A large evaporatively cooled site can draw on the order of a few million litres a day, and a single megawatt of cooling can consume up to around 25 million litres a year. Volumes depend heavily on cooling design and climate. Air-cooled, closed-loop and liquid-cooled sites use far less, and some use recycled or non-potable water to avoid drawing on drinking supply.
Are AI data centres less sustainable?
AI raises power and heat density per rack, which increases the efficiency stakes. It also pushes adoption of liquid cooling, which can lower both energy and water use per unit of compute. The net effect depends on power sourcing and cooling design at each specific site.
Is renewable-powered the same as carbon-neutral?
Not necessarily. Annual certificate matching can leave a site drawing fossil power during hours when renewables are scarce. Hourly, 24/7 carbon-free matching is a stronger standard. Carbon-neutral claims should also state whether they include embodied and supply-chain emissions.
What standards measure data centre sustainability?
PUE and WUE are the core efficiency ratios. Recognised rating and reporting frameworks add verification on top. In Australia the main reference is NABERS for Data Centres, alongside energy-management standards such as ISO 50001 and corporate greenhouse reporting under the National Greenhouse and Energy Reporting scheme.