Australia's main eastern and southern electricity market has begun planning around data centres as a major class of demand rather than a marginal industrial load. The Australian Energy Market Operator expects their grid consumption in the National Electricity Market to rise almost sevenfold over a decade. For investors, the size is important, but the construction of the forecast is more useful than the headline.
AEMO is not promising that every announced server campus will connect, fill and consume its stated capacity. Its method explicitly allows projects to disappear during the connection process and recognises that a connected facility does not draw its maximum load on day one. That converts the forecast into a conditional infrastructure map: data-centre developers need deliverable power, while generators, networks and storage need enough credible demand to justify capital on the right schedule.
Thirty-four terawatt-hours is a forecast, not a booking
AEMO's 2026 Electricity Statement of Opportunities release puts data-centre electricity consumption at about 5 terawatt-hours in 2025-26 and 34 TWh in 2035-36. Their share of electricity supplied through the NEM grid rises from around 3% to approximately 13%. The figures concern the National Electricity Market, not every electricity system in Australia, and describe consumption rather than contracted revenue for any named operator.
The discovery report accurately captures the near-sevenfold increase. Yet 34 divided by 5 is a planning trajectory, not a pre-sold order book. The load will materialise only when sites secure permits, equipment, financing and grid connections, then attract enough computing demand to increase utilisation.
That distinction prevents two opposite errors. Treating all 34 TWh as certain overstates revenue visibility for developers and utilities. Ignoring the forecast because individual campuses remain uncertain understates the aggregate signal AEMO must use to keep the system reliable.
AEMO discounts the development pipeline
The operator says its updated methodology accounts for project attrition during the connection process, the difference between connection capacity and actual electricity use, and gradual demand ramp-up. Supporting materials, including a dedicated data-centre forecasting overview, are listed on AEMO's ESOO publication page.
Those adjustments matter because a development pipeline can be counted in incompatible units. A proposed site's maximum megawatts represent an engineering request. Its annual TWh depend on hours operated and utilisation. A connection agreement is stronger evidence than an announcement, but even a live connection does not establish a steady load. AEMO's conversion tries to bridge those stages without pretending to know which project wins.
The central forecast may still be wrong. Faster processors and cooling can reduce electricity per computing task. Weak customer demand, financing constraints or planning delays can remove projects. The opposite can occur if AI services expand faster or facilities reach high utilisation sooner. The important feature is that attrition and ramp-up are already inside the official estimate; they should not be added again casually as if AEMO had simply summed marketing claims.
Reliability risk arrives through timing
AEMO says forecast reliability gaps are not forecasts of blackouts. They are early signals that continued investment or other action may be needed, particularly beyond 2030. The timing language is crucial. A data centre can be built faster than a major transmission line, dispatchable generator or long-duration storage project. The system can have enough prospective energy in aggregate and still face a gap if enabling assets arrive later than load.
This is a global constraint, not only an Australian narrative. The International Energy Agency reported in April that data-centre electricity use rose 17% in 2025, even as power per AI task improved. It also identified tighter supplies of transformers, gas turbines, advanced chips and other equipment, plus pressure on grid-connection and planning systems. Efficiency can lower unit demand while total use rises because the number and intensity of tasks expand.
For the NEM, generation volume is only one requirement. AEMO also highlights system security and essential services as inverter-based resources take a larger share. New demand that operates continuously may support investment in generation and networks, but reliability depends on location, flexibility and the ability to serve peaks, not annual TWh alone.
Power access changes project economics before utilisation
For a data-centre developer, land and buildings do not earn their planned return if the grid connection arrives late or at a smaller usable capacity. Capital can sit idle before customer utilisation even becomes the main risk. Facilities with firm connections, staged expansion, flexible workloads or credible on-site supply may therefore deserve different economics from projects whose power exists only in an application queue. This is an inference from the infrastructure sequence, not a ranking supplied by AEMO.
Utilities and infrastructure owners face the mirror image. Building too slowly can constrain valuable load and reliability; building ahead of projects that later disappear can leave customers paying for underused assets, depending on regulation and cost recovery. The forecast supports investment, but it does not remove allocation risk. Location-specific network plans and contracted demand remain necessary.
The favorable case is coordination: large users provide long-duration demand, networks expand efficiently, and new generation and storage arrive before reliability gaps. The countercase is a queue of speculative projects competing for scarce connection capacity, with delays raising costs and forcing the forecast lower. Both are compatible with today's 34 TWh central path.
Connections will reveal which projects survive
The evidence that would change the analysis is observable before 2035-36. AEMO's annual forecast revisions will show whether the central path is moving. Signed connection agreements, completed substations, commissioned generation and storage, and disclosed facility utilisation will identify which announcements became load. Faster computing efficiency or greater demand flexibility would reduce the grid capacity needed for a given amount of digital output; sustained high utilisation would push in the other direction.
Investors should therefore treat 13% as a system-planning signal, not a guaranteed market share for data-centre operators. AEMO has made the pipeline consequential enough to enter reliability planning while retaining attrition and ramp-up. The winners will not merely announce the most megawatts. They will align capital, connection and customer use with a power system whose own construction clock is now part of the investment case.