Rethinking nuclear energy projections in the face of modelling illusions

A new paper in Energy Research & Social Science puts a sharper frame around a pattern that has long been visible in energy planning: nuclear power has repeatedly been assumed to grow faster, cheaper and more broadly than it has in practice. The authors describe this as the nuclear energy paradox, a useful label for the gap between the futures long imagined for atomic power and the much flatter reality of global reactor build-out.

The central puzzle is not whether nuclear has value. Existing reactors still provide low-carbon electricity and remain significant in some power systems. The question is why projections have so often built in large-scale nuclear expansion even as global capacity has remained broadly stuck around the 400GW mark and nuclear’s share of global electricity has fallen from 17.5% in 1996 to below 10% in 2023. That disconnect matters because many scenarios have continued to point towards a world with two, three or even four times today’s nuclear capacity, while the current political pledge to triple nuclear power by 2050 would require a dramatic acceleration from today’s baseline, plus replacement of ageing reactors that will retire over the same period.

What makes the paper especially useful is that it treats these forecasts as more than technical outputs. Its authors argue that energy systems are shaped by nuclear imaginaries: shared stories about what a technology is expected to become. In one older version, fast breeder reactors were supposed to make fuel effectively limitless and turn nuclear into a dominant long-term energy source. In a newer version, small modular reactors are presented as factory-built, standardised, cheaper and easier to deploy than large conventional plants. Both narratives have helped sustain expectations, even though neither has been demonstrated at scale.

That matters because scenarios are never just neutral calculations. They are built from assumptions about cost, learning, construction speed, public acceptance, supply chains and institutions. If a model assumes lower capital costs, shorter build times and smooth fleet deployment, it can generate a large nuclear future with little difficulty. But that does not make the result more likely; it simply means the assumptions have been embedded in the answer.

The paper’s discussion of major integrated assessment models is therefore important. MESSAGE, associated with the International Institute for Applied Systems Analysis, and GCAM, the Global Change Analysis Model developed at Pacific Northwest National Laboratory, both aim to explore long-run energy pathways under different constraints. Neither is a pro-nuclear model as such, yet both have helped shape pathways assessed by the Intergovernmental Panel on Climate Change. The concern raised by the paper is that some of the most nuclear-heavy 1.5C pathways rely on inherited model structures and assumptions that may still reflect an older view of nuclear as a backstop technology.

There is, however, a fairer way to read the modelling literature than to dismiss it outright. GCAM has moved away from some of the very low nuclear cost assumptions seen in earlier socioeconomic pathway work, while newer versions of MESSAGEix-GLOBIOM are more transparent and modular than previous generations. REMIND-MAgPIE, from the Potsdam Institute for Climate Impact Research, now includes endogenous learning for technologies such as solar, wind, batteries and electrolysis. That is not trivial. It shows the field has responded to criticism and that the models are not frozen in time.

Even so, structural blind spots remain. Nuclear can be made to look attractive because it is easy to represent in a model as a single large generator with a capital cost, operating cost, fuel cost, lifetime and emissions factor. If the future assumptions include lower capital expense, no serious supply-chain bottlenecks and limited construction delay, the model can allocate a great deal of capacity to it. Yet the hard problems for nuclear are not physics but delivery: finance, regulation, public consent, project management, supply chains and schedule risk.

Hydrogen suffers a related problem. It has real value as an industrial feedstock and will likely have a role in replacing existing fossil hydrogen and serving selected industrial uses. But it becomes too prominent in scenarios when modelling smooths over compression, storage, pipelines, trucking, liquefaction, dispensing, safety systems and utilisation losses. In other words, the model can make hydrogen look like a universal decarbonisation tool when in reality it is likely to remain a niche solution in many end uses.

The opposite tendency appears with renewables, storage and grid flexibility. Solar and wind are not standalone generators; they sit inside a wider system of batteries, transmission, distribution upgrades, forecasting, market design, flexible demand, inverter controls and operational practice. A nuclear plant is easy to model as one big asset. A high-renewables grid is a network of many smaller assets and decisions. That is harder to capture, and when models fail to represent it properly, they can overstate the cost and complexity of renewables integration.

That distinction matters for industrial decarbonisation too. If the modelling underestimates grid-enhancing technologies, it can make centralised firm generation appear more necessary than it really is. If it overstates the cheapness of hydrogen delivery, it can push industries towards hydrogen corridors when direct electrification, heat pumps, grid upgrades and process redesign may be more practical. If it assumes nuclear cost declines without matching evidence from actual project delivery, it risks overstating the technology’s contribution by mid-century.

The paper’s wider message is not that models are useless. It is that they should be read as conditional statements, not forecasts written in stone. When a scenario shows high nuclear, high hydrogen or constrained renewables, that is a picture of what happens under those assumptions, not proof that the real world must follow suit. The risk is that once such pathways are cited by governments, agencies or industry bodies, the assumptions fade from view while the headline result remains.

For that reason, the authors argue for more disciplined scenario work. Models should be checked against historical forecast error, updated more quickly as market evidence changes and stress-tested for feasibility. High-nuclear and high-hydrogen pathways need explicit diagnostics, not just elegant outputs. Grid-enhancing technologies and distribution flexibility should be represented more directly, not treated as generic add-ons. And scenario reporting should make clear which model families are driving the tails, so that policy-makers can see the assumptions behind the storyline.

The larger lesson is straightforward. Energy systems are not only modelled; they are imagined. When imagination gets ahead of delivery, as has often happened with nuclear and hydrogen, caution is warranted. When deployment, learning and cost declines are moving faster than models expect, as has often happened with renewables, storage and grid tools, the models need to catch up.