AEMO's 40% demand forecast rests on a data-centre pipeline that is 36% cancelled

AEMO's 10-year grid outlook, released this week, says data centres will rise from 3% to 13% of Australia's grid power by 2035-36, driving most of a forecast 40% jump in national electricity demand. The same report carries a quieter finding that undercuts the headline: 36% of the data-centre projects on the books in 2025 were already cancelled, and centres operating in early 2026 used just 26% of their contracted grid capacity (AEMO via The Age/Watoday).

The 40% figure is the headline from the Australian Energy Market Operator's Electricity Statement of Opportunities (ESOO), the annual 10-year planning document regulators and network businesses use to set investment baselines. The operator projects National Electricity Market consumption will grow from 176 TWh in 2025-26 to 250 TWh in 2035-36. That figure covers the grid serving the eastern and southern states and excludes "behind the meter" generation, meaning the household and business rooftop solar and batteries that produce power without crossing the grid. Total system change is therefore larger than the headline, but the growth mix is heavily weighted to large industrial loads.

Data centres are the dominant new load. Their electricity consumption is forecast to rise from 5 TWh to 34 TWh over the decade, nearly sevenfold. The driver is the build-out of cloud and AI infrastructure: training runs and inference at scale require continuous, high-density power, and the projects cluster where grid capacity and land are available. The pipeline behind the forecast has roughly doubled in a year. AEMO counts 225 data centres in development for 2026, up from 97 the year before.

The pace has collided with the political and community flashpoint that AI infrastructure has become in Australia. Local opposition, environmental concerns, and state-level disputes over planning approvals have all intensified as the projects have moved from concept to construction. The cancellation rate is one signal of the friction: 36% of the 2025 project list did not proceed, including some that had signed grid-connection agreements. The underutilisation rate is another: operating centres in early 2026 used just 26% of the grid capacity they had contracted. AEMO's planning baseline now has to absorb that gap.

At the same time, the grid is shedding the thermal generation that has historically matched large industrial loads. About 15 GW of coal and gas capacity is expected to exit the system over the outlook period, even as total demand rises sharply. Replacing firm capacity with variable renewables, storage, and demand response is the central planning challenge the report sets out, and the data-centre boom is arriving into the middle of that transition.

The consumer consequence runs through retail pricing. New large loads are typically priced through long-term contracts that can shift network costs onto other users, depending on how regulators treat the cost recovery. The 26% utilisation figure means that ratepayers are already paying for grid capacity that is contracted but unused, and any further build-out will repeat that pattern unless connection terms or cost-allocation rules change. The Australian Energy Regulator and the state-based regulators will have to decide whether speculative projects continue to be able to reserve capacity they may never use, and on what terms.

The ESOO's official release on Tuesday will be the reference point regulators, network businesses, and retailers argue from for the rest of the year. Whether the speculative share of the data-centre pipeline is treated as a planning risk, a regulatory trigger, or a footnote will determine how much of the 40% demand growth shows up in power bills and how much of it is absorbed by projects that never get built.