AEMO projects Australian data centres at 34 TWh by 2036, lifting grid share to 13% as AI investment accelerates
The forecast, roughly sevenfold FY25 consumption, arrives as $51.9 billion in NSW data centre investment advances through state approvals.
Australia's data centres are forecast to consume 34 terawatt-hours of National Electricity Market power by 2035-36, lifting their share of NEM supply from 3% to 13%, the Australian Energy Market Operator said on Monday (2026-08-24). That projection sits sharply at odds with AEMO's own Step Change scenario, published earlier this year, which put data centre consumption at 12.0 TWh by FY30 and 34.5 TWh only by FY50. The newer figure implies the sector reaches the same absolute load roughly 14 years ahead of what the Step Change trajectory indicated. AEMO has not publicly reconciled the two numbers.7,1
Infrastructure decisions, including grid connection timelines, transmission investment, and contracted generation capacity, are priced off planning assumptions, and an apparent 14-year pull-forward in load growth changes the economics of nearly every long-dated contract in the pipeline. Fitch adds a third figure: AI and cloud services projected at 29 TWh, without specifying a target year. Three estimates from credible sources, using different scopes and horizons, leave project developers and grid planners working from incompatible baselines.4,7,1
Oxford Economics Australia modelling, commissioned by AEMO, puts FY25 data centre electricity consumption at 3.9 TWh, with NEM-connected facilities representing 98% of that load. The NEM served a total market of 285.7 TWh in 2025. A 34 TWh data centre load by 2036 would shift roughly 12 percentage points of demand into a sector that registered negligibly five years earlier, against a total NEM projected to reach 385.5 TWh by 2034.1,3
Investment commitments back the accelerated trajectory. The NSW Government announced in March 2026 that 15 data centre projects worth $51.9 billion would advance through its Investment Delivery Authority. NSW already counted 90 operating data centres at that point, and data centre investment had grown at an average of 65% per year over the prior three years. The sector represented 12% of all non-residential building investment in the state — a concentration that gives the load forecast credibility beyond model extrapolation.1
AI is the proximate driver. AEMO's Draft 2026 Integrated System Plan projects business and industrial electricity consumption rising to 253 TWh by 2050, with AI-related loads representing an expanding share of that total. The Sydney Morning Herald reported in June 2026 (2026-06-24) that overall NEM demand could double over 25 years even as household usage nearly halves, driven by rooftop solar and battery uptake. The growth concentrates in commercial and industrial loads. Data centres sit exactly there.4,6
NEM spot prices on Monday (2026-08-24) illustrated the grid's existing imbalances. Victoria cleared at A$97.49/MWh, NSW at A$92.67/MWh, and Queensland at A$83.73/MWh. South Australia settled at negative A$25.52/MWh, reflecting the solar oversupply that periodically pushes the state's prices below zero. Data centres run continuously, adding flat baseload demand rather than contributing to peak-hour generation. That profile becomes more valuable as dispatchable coal capacity exits, and more problematic if connection timelines slip.
Australia's NEM hit a demand record in the fourth quarter of 2025, the same period renewables crossed the 50% supply threshold for the first time, the ABC reported in January 2026 (2026-01-29). But the coal retirement schedule underpinning the energy transition was calibrated to demand curves that predate the AI build-out. Adding a new class of large, always-on commercial loads while removing dispatchable generation creates timing mismatches AEMO has flagged but not resolved.2,1
AEMO's chief executive, speaking at Australian Energy Week in June 2026 (2026-06-11), cited connection arrangements, data-sharing frameworks, and planning structures as areas requiring further work. With $51.9 billion in NSW projects already advancing through government approval channels, the constraint is shifting from capital to connection queue. How fast that queue clears is the variable that separates the 34 TWh-by-2036 forecast from the 34.5 TWh-by-2050 one. Coal retirements do not pause for planning delays.5,1