The environmental impact of a data centre comes mostly from four things: the electricity it draws, the water it uses for cooling, the carbon tied to that power, and the land and materials it consumes. A single hyperscale site can use as much electricity as a small city. The size of that footprint depends on where the power comes from, how the site is cooled, and how hard the servers run.
That is the short answer. The detail matters, because the gap between a well-sited, renewable-powered facility and a poorly-sited one running on fossil grid power is large. This guide walks through each impact, why AI workloads are changing the maths, and what operators are doing to bring the numbers down. It is a neutral explainer. Data Centre Axis tracks the public record on data centre sites; we take no position in any transaction and sell nothing on this page.
How much energy does a data centre use?
Data centres consume electricity around the clock to run servers and to keep them cool. The International Energy Agency estimates data centres used around 415 terawatt-hours in 2024, roughly 1.5 percent of global electricity. That share is now rising fast.
Energy use breaks into two big buckets. The first is the IT load, the servers, storage and network gear doing the actual work. The second is everything supporting them, mainly cooling and power conversion. The efficiency measure that ties these together is Power Usage Effectiveness, or PUE. A PUE of 1.0 would mean every watt went to computing. A modern hyperscale facility often runs near 1.1 to 1.2, while the Uptime Institute's annual survey puts the global average around 1.56, with older enterprise rooms sitting well above that.
The location of the power decides most of the carbon story. The same server draws the same watts whether it sits on a coal-heavy grid or next to a wind farm. What changes is the emissions per kilowatt-hour behind the meter.
How much water do data centres use?
Cooling is the reason data centres use water. Many large sites use evaporative cooling, where water absorbs heat and evaporates, which is efficient on energy but consumes water that does not return to the local system. A large facility can use several million litres per day during hot weather, with reporting hyperscalers averaging around two million litres a day per site.
There are two numbers to separate. Direct water use is what the site draws on-site for cooling. Indirect water use is the water consumed at the power station generating the site's electricity, which is often larger and easier to miss. Water-stressed regions feel both.
This is now a live planning issue in Australia, where several proposed sites sit in catchments under pressure. Some operators are moving to closed-loop or air-assisted cooling that uses little or no make-up water, trading a small energy penalty for near-zero water draw. The right design depends heavily on local climate and water availability.
What are data centre carbon emissions?
Carbon emissions from a data centre fall into three scopes. Scope 1 is direct emissions, mainly from backup diesel generators that run rarely. Scope 2 is the big one for most sites: the emissions tied to purchased grid electricity. Scope 3 covers the supply chain, including the embodied carbon in servers, concrete and steel.
For most operating facilities, Scope 2 dominates. That is why so much of the industry's decarbonisation effort goes into power procurement: signing renewable power purchase agreements, buying green energy, and siting new builds near clean generation. A site on a renewable-heavy grid can have a fraction of the operational carbon of an identical site on a fossil grid.
Embodied carbon is harder to cut and often overlooked. Building a new data centre locks in emissions from manufacturing and construction before a single server switches on. Reusing existing structures, such as a powered shell, avoids some of that upfront cost.
How is AI changing the environmental footprint?
AI is the single biggest change to the energy picture in years. Training and running large models needs dense racks of GPUs, drawing 50 to 150 kilowatts per rack against the 5 to 15 kilowatts of a traditional rack. That density concentrates both power demand and heat in a smaller footprint.
The knock-on effects are real. Higher density usually means liquid cooling, because air alone cannot move that much heat. Liquid cooling can lower energy use per unit of compute, but the sheer scale of AI deployment is pushing total grid demand and water demand up sharply. The International Energy Agency projects data centre electricity use will more than double to around 945 terawatt-hours by 2030, close to 3 percent of the global total, driven largely by AI.
For buyers and developers, this changes what a viable site looks like. Power that arrives in years rather than decades, and a cooling design matched to high density, now decide whether a site can host AI at all.
What makes a data centre sustainable?
A genuinely sustainable data centre works on every impact at once, not just one headline number. The common levers are clean power, efficient cooling, water reduction, heat reuse, and longer hardware life.
The most effective measures tend to be the following. Clean power supply. Renewable PPAs and siting near clean generation cut Scope 2 emissions, the largest source for most sites. Efficient cooling design. Free-air cooling, liquid cooling and higher operating temperatures lower the energy spent on heat removal. Water reduction. Closed-loop and air-assisted systems cut or remove make-up water draw, which matters most in dry regions. Heat reuse. Some facilities pipe waste heat to district heating or nearby industry instead of dumping it. Hardware lifecycle. Running servers longer and recycling them reduces embodied carbon and electronic waste.
Watch for greenwashing. A low PUE on its own says nothing about the carbon behind the power or the water used to cool the site. The credible picture comes from looking at energy, water and carbon together, against the local grid and climate.
How does Australia's grid and water context shape the impact?
Where a site sits decides much of its footprint, and Australia makes that vivid. Grid connection queues, regional water stress and a power mix that varies sharply by state all shape what a given site's real impact will be. A facility in a renewable-rich region with spare water carries a very different footprint from one on a constrained, fossil-heavy grid.
This is why the public record matters. Planning approvals, grid connection data, water allocations and environmental referrals together describe a site's true environmental profile long before it is built. Reading those signals as a system is how you separate a deliverable, lower-impact site from one that only looks good on a brochure.
Frequently asked questions
Do data centres really use as much power as a city?
A large hyperscale site can. Individual hyperscale campuses are often cited in the range of tens to hundreds of megawatts of demand, comparable to a town or small city, and the largest AI campuses now being planned reach a gigawatt or more. Most data centres are far smaller.
Why do data centres use so much water?
The water goes to cooling, usually evaporative cooling that consumes water as it removes heat. Indirect water use at the power station behind the site is often larger than the on-site draw. Closed-loop and air-assisted designs cut this significantly.
Are data centres bad for the environment?
It depends on the site. The impact is real but highly variable. A facility on clean power with efficient, low-water cooling has a far smaller footprint than an identical site on a fossil grid using evaporative cooling. Design and location decide the outcome.
What is PUE?
Power Usage Effectiveness measures how much of a data centre's total power reaches the computing equipment versus overhead like cooling. A value near 1.1 is efficient; the global average is around 1.56 and older sites run higher. PUE measures energy efficiency only, not carbon or water.
Is AI making data centres worse for the environment?
AI raises power and heat density sharply, pushing total energy and water demand up. It also drives adoption of liquid cooling, which can be more efficient per unit of compute. The net effect is more total demand, which is why clean power siting matters more than ever.