AEMO now plans for data centres at 13% of the grid, which changes where they can go

Thirty-four terawatt hours a year, spread evenly across 8,760 hours, works out at about 3.9 gigawatts of continuous draw ^[18]. The size is not really the complaint. AEMO's stated reason for singling out this load is its shape: data centres run at much the same rate through the day and across the seasons, the way a large industrial site does ^[5]. Households and factories hand a planner a trough to work in. A hall of accelerators does not.

The report flags one trap before anyone else can fall into it. Roughly 13 gigawatts of coal and almost 2 gigawatts of gas are due to leave the system within the decade ^[4], but energy measured over a year and capacity measured at an instant are different quantities and do not subtract ^[17]. Comparing them honestly would need the capacity factor of the 9 gigawatts of generation and storage that reached full output in 2025-26 ^[7], and that number is not in the reporting.

There is a second figure inside the 13% that nobody has printed. If 34 terawatt hours is 13% of the market in 2035-36, the market itself is about 262 terawatt hours ^[19]. Run the same arithmetic backwards through the almost sevenfold rise and today's 3% share, and you get roughly 4.9 terawatt hours of data centre load now against a market of about 163 terawatt hours, which means AEMO's share figure only holds if total demand grows on the order of 60% ^[20]. So the 13% is a statement about electrification as much as about racks. If the rest of that growth arrives late, the same 34 terawatt hours becomes a larger slice of a smaller market, and the reliability problem worsens without a single extra server being installed.

Then there is the quality of the input. Developers have cancelled more than a third of the data centre projects AEMO listed a year ago ^[9]. Wires and transformers have no setting for that: the planning cycle is longer than the interval in which a third of a pipeline can disappear, and the same interval could see it refill. Commonwealth Bank puts the national buildout at as much as A$150bn, about $108bn, by 2030 ^[10], and the Clean Energy Finance Corporation has already named the failure mode if connections lag the load: bottlenecks, higher prices for consumers, and more reliance on fossil generation than was planned for ^[11].

The rulebook that would arbitrate any of this is not settled. The federal government is pushing for new data centres to run on renewable energy, and some regional administrations object ^[12]. Anthony Albanese is expected to use Wednesday's national cabinet to tell premiers that federal approval rules will complement state ones rather than override them, according to the Guardian ^[13]. Queensland and the Northern Territory have been the loudest objectors, and a climate expert told that paper Australia has "one shot to get the rules right" ^[14]. Australia's world-first data centre rules required every state to agree, and two refused in July ^[15]. Some builders are not waiting for a connection at all ^[16]. AEMO's own chief executive, Daniel Westerman, has framed the timing plainly: plenty of capacity is coming before the early 2030s, and the wave after 2030 is the one that decides reliability ^[8].