Australia’s artificial-intelligence race has acquired an unexpectedly physical constraint: electricity.12
The Australian Energy Market Operator now expects data centres to draw about 34 terawatt-hours from the National Electricity Market by 2035–36 — roughly 13% of demand, up from about 3% today. ABC’s comparison is deliberately ordinary: that future load would be close to the electricity used by every household in New South Wales and Victoria today.1
That forecast does not prove every proposed facility will be built. It does show why data-centre policy is moving out of the technology department and into energy, water, finance and federalism. The proposed response is to make the new load help pay for the system it needs: new renewable supply, grid costs and controls on water use.13
The thread is not “AI causes blackouts”. AEMO says reliability is not yet under threat, provided new supply arrives on time. The more defensible story is that a private technology boom is forcing governments to decide who carries the infrastructure cost of its physical footprint.1
The forecast changed the scale of the argument
AEMO’s latest outlook considers 225 known data-centre projects at different stages of development. Even after allowing for a high cancellation rate, the operator expects data centres to rise from about 3% of grid demand today to 13% by 2036, or about 34TWh.1
The forecast is a projection, not a construction schedule. It assumes a pipeline, and the final load will depend on which projects secure finance, connections and customers. But the direction is clear enough to become a planning problem: data centres are no longer a marginal commercial load in the national electricity conversation.14
The policy is becoming a “causer pays” argument
ABC reports that the Albanese government expects to settle new national rules requiring data centres to guarantee new renewable supply to cover their energy use. The government has also flagged water conditions and legislation covering wider AI issues, including copyright, security, safety and skills.1
Light Reading’s earlier account described the same policy direction as legal controls on data-centre energy and water use, with operators expected to underwrite power supply and avoid impacts on energy bills.3
The political mechanism is equally important. Queensland has opposed tying data-centre construction to new renewable projects, and the federal government has said it may legislate nationally rather than let projects play jurisdictions against each other.12
The real fight is over who absorbs the next load
The most consequential shift is administrative rather than cinematic. A data-centre proposal can now be assessed as a node in several systems at once: an AI investment, an electricity load, a renewable-power contract, a water user and a federal-state bargaining chip.
That changes the meaning of “growth”. A project can bring capital and computing capacity while also accelerating transmission needs, competing for water and increasing pressure to deliver replacement generation as older coal units retire. The public question becomes less “Should Australia welcome AI?” and more “What must an AI project provide before the rest of the system accepts its demand?”
This is interpretation, not a reported finding about the eventual winners. The answer will depend on the final standard, project location, connection queue, water source and the way costs are passed through.12
AI infrastructure may become a federalism test
The defensible hypothesis is that Australia’s AI infrastructure race will expand federal power over decisions that have traditionally been negotiated through states and local planning. If large data centres can choose among jurisdictions, a national minimum rule becomes a way to prevent regulatory competition from weakening renewable, water or grid conditions.
That does not mean Canberra will control every data-centre decision. States will still influence land, planning, water and local infrastructure. Nor does a national rule guarantee a coherent outcome: a standard can be too weak, too slow or too blunt for different grids and water basins.
The hypothesis would gain weight if the federal legislation overrides state objections, sets common connection and renewable requirements, and is followed by materially similar approvals across jurisdictions. It would weaken if states retain divergent standards, projects relocate to avoid the requirements or the rules are announced but not enforced.
What to watch next
The next signal is not a dramatic AI milestone. It is the rulebook and the queue behind it.
A national cabinet decision, the text of the proposed federal legislation, final renewable and water thresholds, and AEMO’s treatment of projects that do not proceed will show whether the policy is a binding mechanism or a political promise.14
The practical test will be whether renewable projects obtain power-purchase commitments, whether large data centres publish credible water plans, and whether regulators can show that the new load is not simply being socialised through prices or delayed grid upgrades.
What we should watch next
- 01National cabinet decision and publication of the proposed data-centre energy and water framework
- 02Federal legislation setting renewable, connection, water and AI conditions
- 03AEMO updates to data-centre project assumptions and demand forecasts
- 04Data-centre developers announce renewable power-purchase commitments and water plans
- 05Renewable projects reach or fail to reach financial close after new data-centre demand commitments
Australia’s AI race is now an energy-governance test.
AEMO’s forecast makes data-centre demand large enough to reshape electricity planning. The proposed response links compute approvals to renewable supply, water use and cost responsibility, while Queensland’s objection turns the issue into a federalism dispute. The documented thread is from digital growth to physical infrastructure rules. Whether those rules protect households and communities remains the test.
The next AI bottleneck may be a statute, a substation or a water licence.Sources
- 01Data centres to consume nearly as much power as every home in NSW and Victoria by 2036ABC News · 2026-08-25 · VERIFIED↗
- 02Albanese to settle AI data centre energy rules, overriding defiant QueenslandABC News · 2026-08-24 · VERIFIED↗
- 03Australia plans legal controls on data center energy and water useLight Reading · 2026-07-15 · VERIFIED↗
- 04Oxford Economics Australia data-centre energy consumption reportAustralian Energy Market Operator · 2025-01-01 · VERIFIED↗
Threadmap News is an AI-assisted publication operated through a fictional editorial team. Daniel Mercer is the publication’s editorial byline. Named desk members represent specialist research roles within the Threadmap system; they are not separate human contributors or outside authorities.
The two ABC reports are separated by subject: the demand report documents AEMO’s forecast, while the policy report documents the federal-state dispute. They are not treated as independent confirmation of every claim. AEMO is the underlying official forecasting institution; Light Reading is independent specialist reporting on the policy direction. The key documented link is that a large projected load is being met with proposed rules for renewable supply, water and cost allocation. Effects on bills, communities and federalism remain partly unproven.