Seven sites. That’s the number that anchors this whole debate, and it hasn’t changed since the Coalition first named them: Liddell and Mount Piper in New South Wales, Loy Yang in Victoria, Callide and Tarong in Queensland, Northern in South Australia, and Muja in Western Australia. All are current or former coal plant locations. The pitch has always been that you swap the turbine hall and keep the transmission connection, the workforce, the cooling water licence. Neat, on paper. The trouble, as it usually is with nuclear power in this country, is what happens when you put a ruler against the paper.
I’ve sat through enough Senate estimates transcripts and AEMO consultation papers on this to have a settled view, which I’ll declare early: the engineering case for large reactors in the NEM isn’t absurd, but the timeline being spruiked publicly doesn’t match anything a reactor has actually done in a comparable OECD grid this century. That’s not a partisan statement. It’s an argument about units, and units are where this debate keeps going wrong.
What the Coalition has actually said, and what it hasn’t
The Coalition’s policy, as taken to the 2025 federal election, proposed a mix of large-scale reactors at the coal sites and, further out, small modular reactors. The stated ambition was first power in the mid-2030s at the earliest site, with the fleet operating fully public, financed off the Commonwealth balance sheet rather than through a merchant market model. Peter Dutton’s framing at the time leaned on cost figures that the Coalition itself later revised downward from independent modelling, and Treasury’s costing exercise during the campaign arrived at a total considerably higher than the Coalition’s own number. Both sides published spreadsheets. Neither side’s spreadsheet survived contact with the other’s assumptions about capital cost per kilowatt, capacity factor, or discount rate.
What nobody disputes: no enabling legislation exists yet. The Commonwealth ban on nuclear power generation sits in the Australian Radiation Protection and Nuclear Safety Act 1998 and mirrored provisions in the Environment Protection and Biodiversity Conservation Act 1999. Repealing those is a precondition, not a formality, and it would need to survive a Senate that has shown no appetite for it. State-level bans in New South Wales, Victoria and Queensland would also need separate repeal. None of that is under way at pace as of this month.
The timeline problem, measured against actual builds
Here’s where I get pedantic, because the units matter. Plant Vogtle in Georgia, the most recent large reactor project completed in a market structurally similar to ours, took roughly 15 years from the construction licence being issued to the second unit entering commercial service, and its final cost came in near four times the original estimate. The UK’s Hinkley Point C, still under construction on the Somerset coast, has slipped its in-service date repeatedly since first concrete was poured in 2018, and EDF’s own filings now put the cost above £40 billion for two units, up from an initial £18 billion estimate a decade ago.
Australia starts further back than either of those projects did. We have no domestic reactor licensing regime for power generation, no established construction supply chain, and no workforce with recent large-reactor build experience. CSIRO’s GenCost report, produced with AEMO’s input each year, has consistently put a realistic construction lead time for a first-of-a-kind large reactor in Australia at somewhere around 2040, after accounting for the legislative, licensing and site-approval steps that would need to happen first. The Coalition’s mid-2030s target compresses fifteen-plus years of overseas experience into roughly a decade here, starting from a lower base of institutional capability. I don’t think that’s a matter of opinion so much as arithmetic.
The cost comparison nobody quite agrees on
We’ve covered the cost maths in detail before, and it’s worth revisiting rather than re-litigating from scratch: nuclear versus firmed renewables: the cost maths again walks through why GenCost’s levelised cost figures for large nuclear sit well above firmed wind and solar even under generous nuclear assumptions, and why the Coalition’s supporters dispute the discount rate and capacity factor CSIRO uses. GenCost’s 2025 update again put large-scale nuclear’s levelised cost meaningfully above renewables firmed with storage and transmission, even before accounting for construction risk premiums that every large reactor built this century has incurred.
The counter-argument, made reasonably by nuclear advocates including some engineers I respect, is that levelised cost comparisons undercount the value of dispatchable, weather-independent baseload capacity running at high capacity factors for sixty years. That’s a fair point about apples and oranges, and I’ve made space for it before in does Australia actually need nuclear power?. But “the comparison method is imperfect” is a different claim from “the cost will be lower than CSIRO says,” and the Coalition’s public commentary tends to blur the two.
What happens to the coal sites while the argument runs
This is the part that gets less airtime than it should. Liddell is already closed. Eraring’s closure date has itself been a moving target, and we’ve tracked that story in NSW’s energy transition: the coal exit meets a REZ rollout running late, which is relevant here because the Central-West Orana REZ and the broader transmission build are proceeding on their own timeline regardless of what happens in Canberra. Mount Piper’s owner, EnergyAustralia, has flagged its own closure planning independent of any nuclear promise. Loy Yang’s operators in Victoria are working to their own retirement schedules under AEMO’s Electricity Statement of Opportunities forecasts.
The practical question for anyone running one of these plants is whether to bank on a reactor turning up on their footprint sometime in the 2040s, or to plan for what replaces the coal unit in the gap. Every gentailer I’ve spoken with treats the nuclear option as a scenario to note in a risk register, not a plan to build around. That’s not politics. That’s asset managers being asset managers.
Who actually pays, and on what terms
The Coalition’s model proposes public ownership and financing, which sidesteps one genuine problem with nuclear economics: private capital has shown limited appetite for reactor construction risk anywhere in the world without government underwriting, loan guarantees, or a regulated asset base allowing cost recovery during construction. The UK’s Regulated Asset Base model for Sizewell C is essentially an admission that merchant risk doesn’t work for this technology. Fair enough, as a design choice. But it means the Commonwealth balance sheet carries construction overrun risk directly, and Vogtle and Hinkley both suggest that risk is not small.
We’ve written before about who ultimately carries the cost of big transmission and generation builds in who really pays for the transmission build-out, and the same question applies here with more force, because a nuclear build socialises construction risk onto taxpayers from day one rather than through a regulated return mechanism spread across decades of consumer bills. That’s a genuinely different risk allocation to the Capacity Investment Scheme approach, which we’ve detailed in how the Capacity Investment Scheme actually works, where the Commonwealth underwrites a revenue floor rather than the construction cost itself.
Small modular reactors as the fallback argument
When the large-reactor timeline gets challenged, the conversation often pivots to small modular reactors as the nearer-term option. We’ve done the legwork on where that technology actually sits in small modular reactors: the actual state of the technology, and the short version, without repeating a phrase I’ve already worn out on this desk, is that no SMR design has reached commercial operation in an OECD market at the scale being discussed for Australia. NuScale’s flagship US project was cancelled in 2023 after cost estimates rose sharply. That’s not a reason to write the technology off entirely, but it is a reason to treat it as a research and development bet rather than a 2030s delivery plan.
The fair reading, for what it’s worth
I’ll give the nuclear case its due on one point: a country that already runs uranium mines at Olympic Dam and Ranger, and that has run a research reactor at Lucas Heights for decades under ARPANSA oversight, is not starting from zero on nuclear regulation and safety culture. That’s a genuine asset the “we’ve never had nuclear here” line understates. It’s also a much smaller asset than a civil power reactor construction and licensing regime, and conflating the two is doing a lot of work in the pro-nuclear argument at the moment.
My own read, and I hold it loosely because policy positions this far from legislation can change fast, is that the 2030s delivery dates function more as a rhetorical anchor for the ban-repeal argument than as an engineering commitment anyone expects to be held to. Whether that’s honest framing or just what happens to any long-dated infrastructure promise in opposition, I’ll leave to the reader. AEMO’s Integrated System Plan, which models the whole system out to 2050 under its published scenarios, still doesn’t carry nuclear as a committed build in any of its core cases, and that document is where the grid’s actual planning happens, ministerial press conferences notwithstanding.
Agree on the units, and a lot of this argument gets shorter. Overnight capital cost per kilowatt, construction duration measured from licence to commissioning rather than from announcement, and a discount rate that reflects actual sovereign and construction risk rather than an assumed best case. Do that, and the gap between what’s promised and what reactors have delivered elsewhere doesn’t close. It just becomes easier to see.
– Tom Fitzgerald, Baseload & Fuels Correspondent
Photo by Andreas Felske on Unsplash