AEMO’s reliability forecasts have a habit of reading like actuarial tables written by someone who’d rather not be the one telling you the bad news directly. The latest Electricity Statement of Opportunities again flags the mid-2030s as the period where the NEM’s reserve margins get genuinely thin in a couple of regions, assuming committed projects land roughly on schedule. The word “assuming” is doing a lot of work in that sentence, and it always has been.
I’ve sat through enough AEMO briefings now to notice the pattern repeats every year with minor variations: the risk window moves a little, the committed project list grows or shrinks, and the language around “unserved energy” stays carefully calibrated to avoid sounding either complacent or alarmist. Fair enough, that’s their job. Mine is to tell you what the numbers actually imply once you strip the hedging out.
Agree on the units first #
The reliability debate about coal closures gets muddled almost immediately because people argue past each other on what’s actually being measured. Capacity isn’t the same as energy, and neither is the same as reliability in the sense AEMO uses the term, which is a probabilistic measure of expected unserved energy against a standard set under the National Electricity Rules. A gigawatt of coal running at an 85 per cent capacity factor is not equivalent to a gigawatt of battery storage that might run for two to four hours a day, and it’s not equivalent to a gigawatt of wind that might average 35 to 45 per cent depending on the site. None of that is controversial among people who actually run the numbers for a living. It gets lost constantly in public argument because “gigawatt” sounds like a unit of sameness when it absolutely isn’t.
So when someone says coal closures are outpacing replacement capacity, the first question has to be: replacement of what, measured how? Nameplate megawatts replacing nameplate megawatts is the wrong comparison. The right one is firm, dispatchable capability (including storage duration, wind and solar’s contribution at times of peak demand, and interconnector capacity) against the closing plant’s historical contribution during stress events. AEMO’s own Electricity Statement of Opportunities does this calculation properly. Most newspaper columns on the subject don’t.
What’s actually closing and when #
Eraring, still the largest power station in the NEM, has had its retirement date renegotiated more than once, and the current position – effectively holding the plant available into 2027 under arrangements between Origin Energy and the NSW Government – tells you something important about how these transitions actually get managed in practice. Nobody’s waiting for the grid to fail before acting. We covered the mechanics of that specific deal in our piece on how 2025 became 2027 for Eraring, and the short version is that reliability concerns, not sentiment, drove the extension.
Yallourn in Victoria is tracking toward closure later this decade, and EnergyAustralia has been explicit in its own investor materials about the timeline and the replacement assets it’s building to manage the transition, including Tallawarra B. We’ve written about that balancing act in our profile of EnergyAustralia’s countdown. Loy Yang, Bayswater and Vales Point all sit somewhere on their own trajectories, with AGL’s own disclosures setting out indicative dates that have already shifted once or twice in response to market conditions and government pressure, as documented in AGL’s ASX filings.
The pattern across all of these is the same: closure dates move later, not earlier, and they move in response to reliability modelling rather than ideology on either side. That’s worth sitting with for a second, because the public argument tends to assume closures are either being recklessly rushed or stubbornly delayed depending on which side of politics is talking. The actual record is messier and more boring than either narrative – dates slip because the replacement isn’t ready, full stop.
The replacement ledger, honestly totted up #
Batteries have been the standout performer against their own schedule, broadly. The Waratah Super Battery in NSW came online and has already been dispatched during system events, which is more than you can say for a lot of transmission projects on the drawing board; we went through its early performance in our piece on the shock absorber finally earning its keep. Akaysha Energy, Neoen and a growing list of developers have projects either operating or under construction at a pace that would have seemed implausible five years ago.
Gas peakers are a more mixed story. Kurri Kurri eventually burned its first gas after a drawn-out construction period that ran longer than anyone at Snowy Hydro would have liked to admit publicly, a saga we tracked in our status piece on the peaker that took the long way round. The broader gas peaking fleet AEMO’s Integrated System Plan calls for is still well short of fully committed, which matters because gas plays a specific role coal can’t easily be substituted out of: fast-start, flexible generation that can fill genuine multi-hour gaps when wind and solar both underperform simultaneously, something we unpacked properly in our comparison of gas peakers against big batteries.
Transmission is the part of the ledger that worries me most, and I don’t say that lightly. New England REZ, Central-West Orana, VNI West: all of them are either behind original schedule or caught in some form of route or community dispute. You can build all the generation and storage you like, but if the wires to move it aren’t finished, the capacity sits stranded on the wrong side of a constraint. That’s not a hypothetical; it’s happened in smaller pockets of the grid already during periods of high renewable output and low demand.
Where the contrarian case has a point #
Here’s where I’ll part company slightly with the more confident voices on the replacement-is-ahead side of this argument. I think the public reliability conversation underrates how much of the recent calm has been carried by the current coal fleet simply running longer and harder than planned, rather than by the new stuff pulling genuinely ahead of schedule. Eraring staying open to 2027 isn’t evidence the transition is on track; it’s evidence of exactly the opposite, dressed up as a managed outcome because the alternative was politically and operationally worse. I’d rather people called that what it is: a stopgap bought with life extension, not a sign the build-out is winning the race.
That’s not an argument against the transition, and I’d ask readers not to read it that way. The physics of an ageing coal fleet don’t improve with flattery; these units were designed for a different operating pattern than the one they’re now being asked to run, with more frequent cycling and lower average output as rooftop solar eats into midday demand. We went through that minimum demand squeeze properly in our piece on rooftop solar and the midday problem. Running an old thermal unit harder in that cycling pattern tends to increase unplanned outage risk over time, not reduce it, which is precisely the reliability concern everyone claims to be managing.
What the AEMC and AER keep saying #
The AEMC’s reliability settings review work over recent years has consistently pointed to the same conclusion: the current reliability standard is achievable, but the margin for error in meeting it has narrowed as the fleet composition shifts. That’s a polite regulator’s way of saying there isn’t much slack left if a couple of large projects slip at once. The Capacity Investment Scheme, which we covered in detail in our explainer on how the scheme actually works, exists precisely because the market alone wasn’t delivering firm capacity at the pace needed, and governments decided underwriting was cheaper than discovering the gap the hard way during a heatwave.
The AER’s own assessments of network reliability, published through its standard performance reporting, back this up from a different angle: outage frequency and duration metrics haven’t deteriorated dramatically yet, but the modelling behind future years carries wider confidence bands than it used to. Translation: everyone’s reasonably confident about next summer and less confident about the one in 2031.
A Perth aside, and a wider point #
I’ll admit WA’s SWIS gives me a slightly different vantage point on all this, since it’s a grid that can’t borrow reliability from a neighbour the way the NEM states occasionally can over an interconnector – we’ve covered that isolation properly in our piece on the island nobody can send power to. Watching Collie’s transition unfold from this side of the country, the lesson that keeps repeating is that announced closure dates are promises made under current information, and current information changes. That’s not a criticism of any particular operator; it’s just how long-lived infrastructure decisions actually behave when the system around them keeps moving.
The honest position on coal closures and reliability isn’t “everything’s fine” and it isn’t “the lights are about to go out.” It’s that the margin has narrowed, the replacement build is real but uneven across technology types, and the thing propping up the gap right now is old coal plant running past its original use-by date. Whether that’s a sensible bridge or a slow-motion problem depends almost entirely on whether transmission and firm gas capacity catch up before the bridge-builders run out of patience, or budget, or both.
Worth watching which of the three (batteries, gas, or transmission) ends up being the actual constraint when the next genuinely stressed week arrives. My money’s on transmission, but I’ve been wrong about these things before, usually in the direction of underestimating how much can get built in eighteen months when the Capacity Investment Scheme’s money starts actually flowing.
– Tom Fitzgerald, Baseload & Fuels Correspondent
Photo by Gez Xavier Mansfield on Unsplash