The Data Centre Decade: What Australia Is Building, and What It Costs

Australia is in the middle of a build-out it has not finished arguing about. By March this year the country had 162 data centres running, with a combined IT capacity of 1.5 gigawatts. Sitting behind them, approved but not yet built, are another 27 with a combined capacity of at least 3.38 gigawatts. That is more than twice everything running today, and it has already been waved through.

Over the past six months a thread of stories has run through the emergent-tech desk at N.A.N.A., our regional news project, each one a short, in-character take on a single announcement. This is the long version: the same sources, the same figures, laid end to end so the shape of the thing is visible.

How big is the thing we're building?

The global baseline, from the International Energy Agency's Energy and AI: data centres used about 415 terawatt-hours in 2024, roughly 1.5 per cent of the world's electricity, and the IEA's base case has that more than doubling to around 945 terawatt-hours by 2030, about what Japan uses today. The figure to carry through everything below is that a typical AI-focused data centre draws as much power as 100,000 households, and the largest now under construction draw twenty times that. The full set of numbers, including where the power comes from and who feels the growth most, is in the companion explainer: How Much Electricity Does AI Use?

Where Australia fits

Australia's 1.5 gigawatts of operating capacity is small on the world stage. What is not small is the pipeline. The ABC's analysis of approved-but-unbuilt projects, verified by analysts at RenewMap and CBRE, found 27 sites totalling at least 3.38 gigawatts, with names on them like Amazon Web Services, Microsoft, AirTrunk, CDC, NEXTDC, Goodman and Stockland. NEXTDC's S4 in western Sydney alone is planned at 365 megawatts. Growth is concentrated in western Sydney and on Melbourne's outskirts, with some regional sites paired to renewable energy projects.

Then there are the proposals that dwarf even those.

In April, plans emerged for Project Meridien, a data centre south of Broome on Karajarri country, two thousand kilometres north of Perth. It would start at 240 megawatts with a site allowing a full gigawatt, powered by wind, solar and gas, opening around 2032. The Karajarri Traditional Lands Association is a one-third partner. Most Australian data centres run at five to ten megawatts.

In June, IREN announced an 800-megawatt, $10 billion AI data centre beside the Bundey substation in South Australia's Mid North, targeting 2028. The local mayor expected the first water fill to come off the Morgan to Whyalla pipeline, which serves more than 130,000 homes and draws from the Murray. IREN's position is that the facility uses a sealed, air-cooled closed loop losing under one per cent of tank capacity a year. Asked for the exact water figure, the chief executive said he did not yet have it.

In July, Beetaloo Energy proposed a $40 billion, two-gigawatt AI data centre on 185 hectares at Weddell, forty minutes from Darwin. It would need more than six times the combined peak demand of Darwin and Katherine, fed by Beetaloo gas through a new pipeline and a new gas-fired power station. The managing director called it the next wave of the Industrial Revolution and promised town halls. Greenpeace called it a shocking example of unchecked expansion.

Three proposals in four months, and between them more capacity than the entire country operates today.

The three inputs nobody wants to talk about

A data centre is a building that turns electricity into computation and waste heat. To do that it needs three things in quantity: power, water and metal. Each has its own story this year.

Power

Victoria forecasts data centre electricity use in the state to quadruple by 2030. The federal government wants legislation by early 2027 requiring new data centres to underwrite new renewable energy to cover their consumption, pay their own connection costs, be as water-efficient as possible and be built in appropriate locations. The exact standards have not been released. What the ABC analysis made plain is the timing: the rules will arrive after the 27 largest projects have already been approved, and will not apply to them.

Water

Evaporative cooling uses water; closed-loop cooling uses far less but still needs a first fill and periodic top-up. In September, Victoria became the first state to act. Under its new data centre strategy, new facilities must stop cooling with drinking water and use recycled or alternative sources, supply their own renewable power and storage, keep a 150-metre setback from homes and stay away from schools and childcare. Premier Ben Carroll's line was that they must power themselves, protect homes and pay their way. Data centres currently use less than one per cent of Victoria's drinking water; the stated aim is zero.

The industry body, Data Centres Australia, called the rules disappointing and noted they are not applied to other industries. And the rules are not retrospective: the nine applications already approved or before the minister are exempt. New South Wales has no equivalent policy yet.

Metal

CSIRO put a number on the quiet input in May. The average Australian home contains about ninety kilograms of copper. Wind turbines, solar panels, electric vehicles, grid upgrades and data centres all draw on the same metal, and renewable systems can need up to five times more of it than the fossil-fuel plant they replace. Australia holds about ten per cent of the world's copper resources and produces about three per cent of supply. The hopeful figure in the same release: roughly two-thirds of all the copper refined in the past century is still in use somewhere.

And then what?

The other half of any infrastructure story is what happens at the end of the equipment's life, and here the warning comes from a technology we have already rolled out at scale.

One in three Australian homes has rooftop solar and around 300,000 new systems go in each year. Australia generates roughly 60,000 tonnes of panel waste annually, with federal projections of 90,000 tonnes by 2030. About 17 per cent of panels are currently recycled. In August the Australian Academy of Technological Sciences and Engineering recommended a national product stewardship scheme funded by a $10 to $20 fee per panel at installation. Canberra flagged a scheme in 2017; the pilot due mid-year had not started. The counterpoint, from ANU's Professor Andrew Blakers, is that panel waste is about a fifth of one per cent of the national solid waste stream, and that panels are around 90 per cent glass, which is cheap and safe.

The point is not that solar was a mistake. It is that we built the fleet first and are designing the recycling scheme nine years later. Data centre hardware turns over on a three-to-five-year cycle. The accelerated servers that do the AI work are far more material-intensive than a glass panel.

What a sensible regime would ask

None of this is an argument against building. Computation is infrastructure now, and a country that refuses to host it will import it at a markup, with all the same water and power costs paid somewhere else. The argument is about sequence. Victoria's September rules, the federal government's planned 2027 standards and the ATSE's stewardship proposal are all the right questions, asked after the answers were already locked in for the largest projects.

A regime that worked would do four things, and do them before approval rather than after:

• Power: new capacity underwrites new generation and storage, and the connection cost sits with the operator, not the grid. This is the federal proposal; it needs to apply to the pipeline, not just to projects lodged after 2027.
• Water: no potable water for cooling, with the first-fill and top-up figures disclosed at application, not discovered by a mayor. Victoria's rule, applied nationally.
• Materials: a stewardship levy on server hardware at import, on the solar panel model, so the recycling scheme exists before the first refresh cycle rather than nine years after.
• Location: setbacks from homes and schools, and a presumption toward sites co-located with the generation that powers them.

Every one of those already exists somewhere in an Australian policy document. What does not exist is the sequencing. The short-form question that runs through every N.A.N.A. piece on this subject is the right one for a national policy as well: lovely gadget, and then what?