Does GenCost treat nuclear unfairly?

CSIRO prepares an annually updated report on electricity generation technology costs in Australia, GenCost, through a consultative process including a public draft. GenCost’s estimate of the changing capital cost of different technologies over time is a key input to the AEMO Integrated System Plan, the main blueprint for evolving Australia’s largest electricity market. By popular demand, GenCost also includes ‘levelised cost of energy’ (LCOE) calculations as a quick and dirty guide to how the capital costs of different technologies translate to their relative costs in practice.

Levelised cost is not the be-all-end-all, but tries to take all the costs of financing, building and running a new generator over its life, all the energy it will produce in that time, discount future costs and benefits to reflect the time value of money, and convert that to a single cost per unit of energy. LCOE makes for handy comparisons, but it is a pretty crude and basic tool.

Similar exercises have been run for many years and always attract gripes from people who feel their favourite technology has been done dirty. A dozen years ago it was wind and solar fans saying that future costs were wildly overestimated by the 2012 Australian Energy Technology Assessment (and as it turned out they were right – costs fell much faster than projected by anyone but Greenpeace). Lately it is nuclear fans who complain the most. GenCost concludes that the costs of nuclear electricity are very high in Australia compared to other options.

A recent AFR article by John Kehoe rounds up the most common objections:

1. Once built, nuclear plants can operate for much longer than the 30 year period over which technologies are compared in CSIRO’s calculations.
2. Nuclear plants are capable of operating nearly all the time, rather than for as little as half the time as considered by CSIRO.
3. Nuclear generates electricity all the time, rather than only when weather allows, so it can’t be compared like for like with fully variable renewables.

?As it turns out, these issues are all considered pretty thoroughly in GenCost, and the answers there are quite reasonable, if incomplete. The short version is that plant life beyond 30 years doesn’t make much financial difference to real-world investment decisions; that real-world electricity market conditions increasingly require ‘baseload’ generators to either operate less often or accept very low prices for their power; and that the costs of flexibility apply to baseload generators too, are being deeply considered, and are manageable if done well.

Operating life

Nuclear fans say that reactors can run for perhaps 80 years, while renewables have to be replaced after 20 or 30 years – and that GenCost tilts against nuclear by only calculating levelised cost over a 30 year period. But plant life beyond 30 years does not make a huge difference to a financial decision about investment in a new electricity generator.

A Levelised Cost Of Energy calculation (or any more complex and nuanced investment decision calculations) involves discounting future costs and benefits of a power plant to consider the best allocation of money now. Money has a time value. A dollar of revenue or cost today is worth more than the same in a year’s time, and a lot more than the promise of a dollar in 80 years. That’s partly because inflation is a thing (as if we needed any reminder these days!), and partly because people loaning, borrowing or investing dollars want a competitive annual return from them. ?

At anything like the (reasonable for commercial purposes!) 5.99% discount rate used in GenCost, the potential of a nuclear power plant to operate beyond 30 years does not much change its levelised cost. The value today of a dollar earned in 80 years’ time would be less than one cent. With the most generous other assumptions in GenCost (especially a high utilisation rate, see below), costing nuclear over 30 years produces a levelised cost of about $122 per megawatt hour. Extending that all the way to 80 years (and assuming no major refits or reinvestment of any sort) cuts the levelised cost to $110. Not much difference!

A lower discount rate that gave greater weight to long-term benefits could make nuclear look cheaper – low interest loans make everything cheaper, though they’ve been harder to come by lately as governments’ own cost of borrowing has gone up. On the other hand a lower discount rate would also increase the present value of future decommissioning costs for a nuclear facility, which are large eventually but are so far off that they are (properly!) ignored in conventional analysis like CSIRO’s. A lower discount rate would also imply much more aggressive action on emissions reduction now to minimise the enormous costs of climate change later (or indeed the lower, but still very high, costs of engineered negative emissions to recapture carbon emitted now).

Meanwhile the costs of replacing renewables in future are also easily exaggerated. The sites for wind and solar farms and their network connections can continue to be used whenever the panels or turbines themselves need to be replaced, avoiding many of the initial costs of establishing a new renewables facility. And the capital equipment that does need to be replaced will continue to decline in cost with further global deployment: future wind and solar plants will be considerably cheaper to build than recent ones.

For instance, a large scale solar plant built in Australia in 2023 had a capital cost of $1,526 per kilowatt. At the current rate of global deployment, by the time that plant needs to be replaced in 2053, CSIRO project that the capital cost will have fallen to $784 per kilowatt. If deployment of all technologies speeds up to the level needed for global net zero emissions by 2050, the large PV plant capital cost falls to $516 per kilowatt by 2053 – a reduction of two thirds, ignoring savings from site reuse.

The learning rates for solar and wind (how much cheaper the next installed unit gets as a function of the cumulative total of global installations) are extremely well demonstrated through the last few decades of cost and deployment data. We can be very confident in them; what we don't know for sure is how fast the world will deploy different technologies, giving the learning rate a chance to work; though the signs are that deployment will remain very rapid. (Learning rates for nuclear are tougher to be confident about since there are fewer, bigger installations, but they turned negative in the 1960s – new plants have grown more expensive than old plants since then, though there are intense efforts to turn that around.)

Renewables plant life may be able to be extended further than the 20-30 years assumed for different technologies in exercises like GenCost. For example a Singaporean startup is commercialising a technology for rejuvenating solar panels in situ at a cost much lower than replacing them. We’ll see how that goes in practice; certainly responsible management of end-of-life solar panels is essential, and the product stewardship scheme being developed to ensure this should be a high priority.

All up, the conventional levelised cost of energy calculation deals adequately with plant life, and an approach that was more complex or put a lower time value on money would be unlikely to make nuclear look very much better.

Utilisation rate

CSIRO critics are aggrieved that GenCost considers nuclear operating at a capacity as low as half its potential, when it could run full-tilt most of the time. But GenCost considers this very reasonably given what is actually happening in electricity markets in Australia and around the world.

The utilisation rate or capacity factor are measures of how much of its theoretical potential energy output a generator will actually supply – the fraction of their rated capacity at which they operate on average. (Typical usage is that capacity factor is the estimate used in modelling, while utilisation rate is what’s achieved in practice.) For instance, a large scale solar plant with 100 megawatts of maximum output will reach around that level for a few hours a day, operate at a lower output for a few more hours, and supply nothing at night; the utilisation rate might be around 30%.

Another example is a gas peaker plant, also with 100MW maximum output. In principle the peaker could operate all the time, but it is intended for flexible operation just when market prices support it. A reasonable assumed utilisation rate might be anything from 5% to 20%.

Nuclear power plants are physically capable of operating at full capacity nearly all the time, if market demand (and prices) and their own costs justify it. The more operating hours a nuclear plant can run, the more it can spread out its high capital costs and offer a lower unit price for its energy while still repaying its investors.

The operating costs of nuclear (once built) are low but not zero – fuel costs vary with output and a lot of workers are needed to keep it running. Nuclear can ramp up and down if it has to, but there is wear and tear associated with this; most designs would really rather run flat out. Of course, as with any energy technology, sometimes things can go wrong and the plant can’t run fully. For instance, French nuclear reactors were throttled down at times in 2022 and 2023 because severe droughts limited the available water for cooling them.

So nuclear could run nearly all the time, and would really like to. However, there is a major problem with assuming that it will! The energy system is ever more full of large scale renewables and rooftop solar. They are not going away, and they change the game for physically or financially inflexible generators like coal and nuclear.

Large scale renewables have a short run marginal cost of generation of zero: once they are built it costs them nothing more to send power out than to not send it out. They can afford to underbid coal or nuclear and still be cashflow-positive. And while they do need to recover their capital costs and make a return, they can do so at a quite low average price for their power – perhaps $50-60/MWh for solar built in 2023, perhaps $25-30/MWh for solar built in 2040.

Meanwhile rooftop solar is devouring electricity demand before it is even visible to the wholesale electricity market. The costs are very attractive to households and businesses for meeting their own needs, even though the level of public support through the RET has been trailing off and will be gone in five years.

Wind is plentiful sometimes and scarce at others, though it can be predicted well. Solar is also variable but even more predictable and regular. The result is that while there is plenty of renewable energy about on average, at times there is none and in the middle of the day especially there is lots of supply and ever less visible demand.

Spot electricity prices are now often zero or even negative in the middle of the day. They go below zero in large part because coal generators would rather pay for the privilege of continuing to run, rather than bear the wear-and-tear to their equipment from ramping up and down too much. Minimum demand for wholesale market electricity – the famous ‘base load’ that can be met by always-on generators – is getting close to zero in South Australia on some days of the year, and the trend is in the same direction everywhere.

Transmission lines to better connect different regions can help smooth these surpluses out, as can storage. That helps integrate renewables, but it would also create a friendlier market for nuclear to sell to more customers over wider geography and time (though the challenge of competing on cost would remain).

What’s happening today with coal generators is pretty indicative. Coal generators are also physically capable of running at full capacity almost all the time, but they don’t because they are facing these market trends. Historically Australia’s coal generators ran at around 60% utilisation rates. Black coal generators in NSW have recently run at utilisation rates from 55% to 80%. Brown coal in Victoria, with much lower fuel costs, managed 83-91%. They achieved these rates by accepting electricity prices often much lower than they’d like. They expect this situation to worsen, which is why they are all planning their exits.

In other markets with nuclear today it can have as much difficulty as other relatively inflexible technologies in running as often as it would like. In competitive markets in the United States, existing nuclear has been struggling to keep going despite already sunk capital costs. Competition has been huge, not just from renewables but also from gas generators, given their greater flexibility and – at extraordinarily low US gas prices – only moderately higher operating costs. The US Inflation Reduction Act included subsidies to existing nuclear plants to help them survive.

Given all that, CSIRO very reasonably looks at a range of potential utilisation rates for nuclear and other dispatchable technologies. They consider a top rate of 89% because these plants are capable of it. They consider a minimum rate of 53% because markets seem no more likely in future to support very high utilisation of inflexible generators than they are today.

What difference does this make? The levelised cost of nuclear electricity is certainly lower with a high utilisation rate than a low one, but still not that low. For example, with GenCost assumptions for nuclear delivered in 2040, including a capital cost much lower than that achieved by actual recent Western nuclear plants, a 53% utilisation rate produces a levelised cost of $196/MWh. 89% utilisation would reduce that to $122/MWh. That lower end is still around five times the cost of unfirmed large scale solar in the same time period. That implies that nuclear would still struggle to achieve a high utilisation rate in a competitive electricity market with large scale renewables and rooftop solar.

Nuclear could still be considered for a firming role, rather than bulk baseload generation. The costs of other firming options would have to be very, very high before nuclear was competitive in that role; though there are interesting debates over the best options to meet the last five per cent or so of annual energy needs in a very heavily renewable grid, and nuclear is one of the options that could be considered there.

Now of course energy market design, complementary policies or government contracts could protect and prioritise nuclear to operate at higher rates. One way to achieve this would be to forcibly direct large scale renewables, and household and business solar systems, to turn off in order to create sufficient space in the wholesale electricity market for nuclear to have a clear run.

Another approach, perhaps more plausible but quite expensive, would be to provide a public financial guarantee of minimum electricity prices for nuclear. The UK took this approach with the contract underpinning Hinkley Point C. Once that plant is built (in 2008 it was promised for 2017 but is currently aiming for 2026), the UK government will top up the electricity prices it receives as necessary to ensure the generator makes money and can keep dispatching power.

The public expense of such guarantees can be quite substantial if the benefitting generator has costs much higher than actual electricity prices. Hinkley Point C’s government contract guarantees it ￡92.50/MWh in 2012 money, to be adjusted upward with inflation for the next few decades. That is currently equivalent to AUD$260/MWh per megawatt hour, to be inflation adjusted for the next few decades.

On balance, the CSIRO approach of considering both low and high utilisation rates is reasonable. Nuclear faces a serious challenge achieving high rates in practice. The trade-off is either low utilisation and a high unit cost of energy; or high utilisation achieved by accepting prices well below those needed to recover project costs (requiring investors to lose their money or receive a large public subsidy).

Apples to apples comparisons

On what basis can you compare a dispatchable energy resource, that operates when you want it, with a variable one that operates only when it can? This is both a complaint of GenCost critics, and a thoroughly studied question extensively considered in GenCost itself. CSIRO discuss the issues and make clear distinctions between fully variable renewables, firmed renewables, high-capacity flexible power and peaking generators.

Electricity supply has to equal demand at (nearly) all times. The availability of variable renewables doesn’t necessarily coincide with demand; making them match better takes actions that will have costs. But the situation is nuanced!

Renewable supply can be made to match demand in many ways.

Some demand can shift to when renewables are most available; for example water heating and car charging are best timed for the middle of the day amidst solar abundance.

Storage can also shift energy from high supply to low supply periods. Batteries are fantastic for this where the need is regular (like daily peak demand in the late afternoon and early evening) and the gap to be filled lasts up to 4 hours. 8 hour batteries are becoming economically plausible as battery costs decline rapidly.

Pumped hydro can potentially offer larger volumes of cheaper storage for more seasonal needs, though delivery costs have been much higher at Snowy 2.0 than initially hoped, and batteries may consume large parts of the market pumped hydro aims at.

Transmission lines are important – not only are they needed to open up new renewable energy zones, but they can shift energy across the country to make supply match demand. The wind in Tasmania is rarely correlated with the wind in Queensland, and a more interconnected grid benefits from that diversity.

Gas peakers are also extremely useful, and can jump into action whenever other resources are not enough. Peakers are expensive to run but cheap to build. That makes them the best bet to fill gaps in supply that are too infrequent and extended for storage to fill economically; peakers are like an insurance policy with a low premium and a high excess.

Demand shift can be free, but all those other flexibility resources have costs. The cost of an adequate mix can be higher or lower depending on how well we do it, but it has to be taken into account.

On the other hand dispatchable but inflexible generators (like coal or nuclear) also don’t necessarily match demand! Demand varies up and down every day and over the seasons. These generators can turn down if they absolutely have to, at considerable cost to themselves, but they’d really rather not. They also benefit from flexibility elsewhere.

Demand can be shifted to make life easier for inflexible generators. That’s why electric resistance water heaters used to be automatically switched on in the middle of the night – to help keep coal generators going. Transmission can share inflexible energy supply to areas of higher demand, as it does for coal power right now. Energy storage can help inflexible generators realise more value for their output by shifting to periods of more demand. And gas peakers can help bridge the remaining gap between the steady demand that inflexible generators would prefer to service, and the peaks that occur every day and seasonally.

There is no sense in counting the flexibility costs for integrating renewables but not the costs of integrating inflexibles. A sensible exercise considers the lot. And as CSIRO make clear, these issues are much better addressed through a system planning exercise (using as an input the capital cost estimates that are the core work of GenCost) rather than trying to compare generic new assets on their own.

CSIRO have offered several simplified methods over the last few years for estimating the additional unit costs of digesting variable renewables. But they always say: don’t rely on this for anything more than the crudest comparison, do a system plan.

System planning exercises try to account for the specific geography, existing asset base, demand profile and weather conditions that an electricity system actually faces, and assesses the best mix of different new assets and technologies to meet that demand with sufficient reliability at least cost.

The Australian Energy Market Operator’s Integrated System Plan is one version of this. It has to work within existing law and policy, so it is not able to consider nuclear while it remains illegal. The ISP does work through the minimum cost configuration of other assets and provides a plausible, manageable-cost pathway that is very heavily renewable, backed by gas and fully reliable.

The ISP could certainly do better on cost – for instance with better use of demand side resources. If it wasn’t required by State policy to include offshore wind it almost certainly wouldn’t deploy any, for cost reasons, but would make more use of onshore renewables, big batteries and gas peakers instead.

The Net Zero Australia Project (by energy system experts at the University of Melbourne, University of Queensland and Princeton University) was a similar exercise on a much larger scale, modelling the decarbonisation of Australia’s exports as well as the domestic economy. NZA was also imperfect, but did consider nuclear. While the same model and some of the same people found a significant supporting role for nuclear in the United States in the Net Zero America project, in Australian conditions nuclear didn’t play any role unless its capital costs fell by a third below the lowest level NZA actually thought plausible, and if onshore renewables were sharply constrained. Even then nuclear only accounted for a small share of total generation.

On balance, CSIRO’s approach in GenCost is reasonable and upfront about its limitations. There is no reason to think that a more nuanced system planning approach would show a dramatically different result for nuclear.

The bottom line

Anybody can put a levelised cost formula into Excel and torture the assumptions until it delivers the result they want. That’s grist for Twitter warriors, but not much use to investors or system planners. They need credible inputs on which they can make their own judgments and specific plans. The meat in CSIRO GenCost is the capital cost estimates, and they are foundational – but also subject to regular revision in the light of experience. That regular revision is essential; there is a long history of similar exercises overestimating the future costs of technologies with strong learning rates, but also the recent surprises from the interest rate hikes and inflationary pressures that followed the pandemic.

By contrast CSIRO’s levelised cost estimates are a quick guide for the simplest surface-level analysis. For all that they represent very defensible judgments by an experienced team, tested through a very open and iterative process. Nuclear fans can relax about the referee and concentrate on solving economic challenges.