Edition 11 · 9 September 2026 · 5 min read
Last edition I said I'd read AEMO's Electricity Statement of Opportunities next to the AER's market report rather than on its own. It landed in late August, and the line that stopped me isn't in the reliability chapter at all. It's a short admission about how much of AEMO's own data centre pipeline has already evaporated.
The Take: the data centre queue is not a demand forecast
AEMO built its forecasts around 225 known data centre projects. The same section carries the caveat: more than 40% of data centre projects since 2025 have either dropped out or regressed in connection status. That's the market operator telling you, in its own report, that two in five entrants to the pipeline underpinning one of the most-quoted demand numbers in the market have gone backwards.
The headline projection is still substantial. Data centre consumption is forecast to grow roughly seven-fold over the decade, from about 5 TWh today to around 34 TWh, and from roughly 3% of operational demand to about 13%. I'm not going to pretend that isn't real growth.
But there's a second figure next to it that gets almost no airtime. AEMO expects mature data centres to draw only about 50% of their network connection capacity, running close to flat around the clock rather than peaking with everyone else. It attributes that assumption to the IEA's Energy and AI work. So the demand line rests on a pipeline where two in five entrants have already gone backwards, and the ones that survive are expected to use half of what they booked.
Connection capacity isn't free, and that's where this stops being an interesting statistic. Networks plan and build against booked capacity. When a project withdraws, the augmentation doesn't withdraw with it, and the cost of it sits in a regulated asset base that everyone connected to that network pays down. The gap between a queue and a forecast is a question about who funds the difference.
Here's what I'd actually do with it. Next time a network or a retailer justifies an increase by pointing at data centre growth, ask two questions. First, what's the connection-status breakdown of the projects inside that number, because a queue position and a committed project are not the same asset. Second, what utilisation factor is assumed against nameplate. AEMO has now published its own answer to the second one, and it's roughly half. If you're modelling network cost past 2028, use that figure rather than nameplate, and if the model in front of you runs on nameplate, you've found something worth arguing about before you sign anything.
Quick Hits
Britain's price cap rose 4%, and the flat electricity line is a tax cut in disguise. Ofgem set the cap at £1,723 a year for a typical dual-fuel direct debit household from 1 October, up about £60. Gas went up 8%. Electricity looks broadly stable, but only because government removed VAT from electricity bills for October through March. Without it, bills would be around £45 higher, and that relief unwinds in the spring. So what: a flat headline rate is not a flat cost. If you benchmark a portfolio across markets, strip the policy effects out before you call anything a saving, because the underlying commodity move is in the wrong direction.
The AEMC is consulting on writing down gas network assets, and submissions close 8 October. The draft determination on gas networks in transition proposes a 20-year outlook alongside the existing five-year framework, new transparency obligations on service providers, tighter alignment between cost recovery and expected network usage, and provisions to reduce asset values where underutilisation occurs. A final determination is expected in December. So what: this is the mechanism that decides whether a shrinking gas customer base carries a rising cost per unit or whether networks absorb the write-down. If you hold gas sites, it's one of the few open processes where a submission from a buyer still counts.
Mid-scale solar between 100 kW and 1 MW moves to upfront certificates from 1 October. The Clean Energy Regulator has confirmed the Small-scale Renewable Energy Scheme is expanding its eligibility ceiling from 100 kW to 1 MW, intended to apply to systems installed from 1 October, subject to the regulations being in place. Systems in that band become eligible to create STCs rather than accruing LGCs annually. So what: deemed upfront certificates change the payback on the most common commercial rooftop size, sometimes materially. The catch is in that subordinate clause. The regulations aren't made yet, so anyone signing on the new treatment before 1 October is carrying real timing risk.
From the blog
If the Take has you questioning a network charge, our explainer on the AEMC's network regulation review covers how those charges get set in the first place, and which parts of the framework are open. Read it here.
The Sign-Off
One more thing from the ESOO worth holding onto. AEMO now flags winter as the growing stress season rather than summer, as heating electrifies across the eastern states. Under its committed and anticipated assessment, the reliability standard is forecast to be exceeded from 2030-31 in New South Wales and Victoria. That's a long way out and forecasts at that range are soft, but the seasonal shift underneath it isn't. Almost every demand response and load-shifting business case I've seen was built on a summer afternoon.
So the question I'm sitting with: if the stress moves to winter mornings, how much of the flexibility we've spent a decade building is pointed at the wrong hours? My honest read is that a good chunk of it is, and that the fix is mostly commercial rather than physical. The assets can respond whenever you ask them to. The contracts they sit under are the part written around summer.
If you've reworked a flexibility case for winter and the numbers moved, I'd like to hear it. Just reply.
Until next fortnight,
Cohen
Find me on LinkedIn.
Get Behind the Meter in your inbox every fortnight. Subscribe →