The real energy and water cost of AI data centres
A single data centre proposed for western Sydney has, at various planning stages, been described as needing enough electricity to power a small city. That’s not marketing copy, it’s roughly the scale some of these facilities now operate at, and it’s worth sitting with for a second before we get to the “but what does this mean for me” part. The honest answer is: more than most of us have been told, and less catastrophic than some of the louder warnings suggest. Our take sits somewhere in the uncomfortable middle, which is usually where the useful answers live.
We’ve watched the AI data centres energy cost conversation shift over the past couple of years from a niche concern among energy analysts to something that shows up in mainstream coverage most weeks. Good. It deserves the attention. But a lot of the commentary either treats every new facility as an environmental emergency or waves the whole thing away as the cost of progress, and neither framing helps an ordinary reader figure out what’s actually at stake for their power bill, their water supply or their suburb.
What a data centre actually uses
Strip away the jargon and a data centre is a very large shed full of servers, each one a specialised computer, running more or less constantly. Training a large AI model, the process of feeding it enormous amounts of data so it learns patterns, is energy-intensive in short, brutal bursts. Running that model afterwards so it can answer your questions, called inference, is a smaller load per query but it happens billions of times a day across the planet, so the total adds up fast.
All of that computing generates heat, and heat has to go somewhere. Traditional data centres used air conditioning. Many AI-focused facilities now use liquid cooling, sometimes evaporative systems that consume water directly, because packing more processing power into less space makes air cooling impractical. That’s the second half of the equation nobody mentioned five years ago: these buildings don’t just want electricity, some of them want water, and in a country that knows drought cycles intimately, that detail matters more here than it might in, say, Finland.
Short version: it’s a power problem and, depending on the design, sometimes a water problem too.

Why Australia is in the conversation
We’re not a global hub the way Virginia or Singapore are, but we’re not a bystander either. Sydney and Melbourne both have clusters of existing and proposed data centre capacity, and the AI boom has accelerated interest from global operators looking at Australian sites partly because of our grid stability and partly because of proximity to Asian markets. Regional sites have been floated too, chasing cheaper land and, in some proposals, access to renewable generation that doesn’t have anywhere else to go.
The Australian Energy Market Operator publishes forecasts that increasingly flag data centre demand as a meaningful new load on the grid, not a rounding error. That’s a change from a decade ago when data centres were a background hum in the electricity market rather than a headline item in planning documents.
Dry one-liner: we used to worry about air conditioners on hot days blowing the grid; now we’re adding buildings that run like that, every day, all year.
The water question locals are asking
This is where we’ll admit to a genuinely mixed view. Some data centre operators now use closed-loop cooling systems that recycle the same water repeatedly with minimal fresh input, and some are built specifically to avoid drinking-water sources altogether, drawing on recycled or non-potable supply instead. Others, particularly older or less efficient designs, draw meaningfully on local water utilities. The difference between those two approaches is enormous, and it rarely makes it into a press release headline.
Residents near proposed sites in outer Melbourne and western Sydney have raised exactly this question at council consultation sessions, and fair enough. If you live near a facility that’s going to draw water during a summer when your own council is asking you to limit garden use, you want a straight answer about which category that facility falls into, not a corporate sustainability statement full of soft language.
Our mildly contrarian view: we think local water use at individual sites gets more airtime than it deserves relative to electricity demand, because it’s the more visceral, easier-to-picture issue. Electricity use is the bigger structural problem for the grid as a whole, water is mostly a site-specific planning matter that a decent approval condition can usually manage.
Who actually pays for this
Here’s the part that should matter most to someone reading this on their phone in Dandenong or Dubbo: do data centres push up household power bills? The honest answer, based on how the National Electricity Market works, is indirectly, and only at the margins, for now. Large new loads can affect wholesale electricity prices and can justify network investment that eventually flows into tariffs, but a single new facility isn’t going to show up as a line item on your retailer’s bill.
The bigger question is whether enough new generation and transmission gets built fast enough to absorb this new demand without squeezing everyone else, households included. That’s a planning and investment problem more than a data-centre problem specifically, but the two are now tangled together in a way they weren’t a few years back.
We’d also gently push back on the framing that your Netflix habit and this AI boom are morally equivalent data centre users. They’re not close. Video streaming is a mature, heavily optimised workload. Training frontier AI models is a comparatively new and much less efficient one, and the industry itself is still working out how to tame that.
The efficiency argument nobody trusts yet
Every major chip maker and cloud provider will tell you their newest hardware delivers dramatically more computing per watt than the previous generation, and that’s often true in a narrow, lab-tested sense. The practical problem is what economists call the rebound effect: when something gets cheaper or more efficient per unit, people tend to use a lot more of it, so total consumption climbs anyway. We’ve seen this exact pattern before with cloud computing generally, and with data storage specifically, our own explainer on free cloud storage limits touches on how usage habits balloon once friction disappears.
Chip efficiency gains are real and worth celebrating, genuinely. But treating them as a solution to the AI data centres energy cost problem, rather than a factor that slows its growth rate, is wishful thinking dressed up as an engineering achievement.
What regulators here are actually doing
Australia doesn’t yet have a dedicated national regulatory framework specifically for AI data centre energy or water consumption, the way some European jurisdictions are moving towards. What we have instead is a patchwork: state planning approvals that increasingly ask proponents to detail water sourcing and cooling design, and the existing energy market rules that the Australian Energy Market Operator and state regulators apply to any large new grid connection, data centre or otherwise.
That’s arguably sensible for now rather than a gap to panic about. Bespoke national rules written in a hurry tend to age badly once the technology moves, and this technology is still moving fast. We’d rather see sharper disclosure requirements, basically forcing operators to publish water and energy intensity per unit of compute, than a blunt cap that might just push the next facility to a jurisdiction with weaker scrutiny.
If you’re curious how this AI boom connects to the tools actually running on these servers day to day, our piece on why AI chatbots make things up and how to catch it is a useful companion read, since a lot of that inference workload is precisely what’s humming away in these buildings right now.
What this means for your own tech choices
None of this means you should feel guilty asking an AI assistant to summarise a council newsletter or draft an email. The per-query footprint of a single chatbot interaction is small; it’s the aggregate, billions of times over, that matters at a grid level, and that’s an industry and policy problem, not an individual guilt trip. We’d say the same about streaming video or, further back, about leaving a desktop PC running overnight, which was the equivalent household-tech energy panic of the early 2000s.
If you want a genuinely useful household-level lever, it’s less about which AI tool you use and more about the devices doing the asking. A laptop or phone replaced every two years instead of five has its own embedded energy cost, separate from anything happening in a data centre, and our guide on how long a smartphone should last before you upgrade is a reasonable place to start thinking about that trade-off. Pairing an efficient home setup with something like a smart plug that actually shows you real consumption, which we cover in our piece on smart plugs and power monitoring, at least gives you visibility over your own slice of the problem, even if it’s a tiny one next to a hyperscale facility.
Our verdict on the AI data centres energy cost debate
We reckon the loudest voices on both sides are overcorrecting. The AI industry’s efficiency claims deserve scepticism, not dismissal, and the doom framing that treats every new data centre as an ecological crisis skips past the fact that well-designed, well-sited facilities with proper water and energy disclosure can be managed sensibly through existing planning and market mechanisms, the same way we’ve managed other large industrial loads for decades.
What we’d actually like to see is boring and unglamorous: mandatory, standardised disclosure of water and energy intensity per facility, published the way the Australian Energy Market Operator already publishes grid demand data, so households, councils and journalists aren’t relying on corporate sustainability reports to understand what’s being built next door. That’s not a revolutionary demand. It’s just transparency, applied to a technology that’s grown faster than our instinct to ask basic questions about it.
The data behind grid-wide generation and demand forecasting is tracked in detail by bodies like the Clean Energy Regulator, and that’s a better place to watch the real trend lines than any single company’s press release.
— Dale Whitfield, Founding Editor, Tech Geek
Featured image: Photo:Alexis Lê-Quôc from New York, United States, CC BY-SA 2.0.




