What the commitments actually require
At COP28 a group of governments committed to pursue a tripling of nuclear capacity from 2020 levels by 2050. Global operable capacity in 2020 was roughly 370 GWe, so tripling implies something in the order of 1,100 to 1,200 GWe. The World Nuclear Association goes further: aggregating stated national targets gives a 2050 fleet of about 1,446 GWe, materially above the headline tripling benchmark. As of late 2025 the world had 413.2 GWe operable and 78.7 GWe under construction.
Those figures understate the build task, because a 2050 fleet target describes an end state and says nothing on its own about the reactors that will close along the way. Depending on how far lifetime extensions are pursued, somewhere between 81 GWe and 156 GWe of existing capacity retires before mid-century, and the WNA estimates that around 1,100 GWe of new capacity must be built to reach 1,446 GWe once replacement is included. Separately, the WNA’s published connection pathway requires new capacity to reach the grid at an average of about 40.6 GWe a year, rising from around 14 GWe a year in the late 2020s to more than 65 GWe a year by the late 2040s, and this pathway forms the delivery backbone of our model.
How the model works
We converted that connection trajectory into a construction spend profile. Each year’s connection requirement is assigned to a construction cohort of a defined duration, and annual spend is the sum of every cohort active in that year. Projects already under construction at the start of 2026 enter as remaining spend only, and are assumed to deliver all grid connections to 2030, since nothing starting after 2025 can credibly connect before about 2031.
Construction duration is the first-order variable, because it governs both how long capital is committed and how heavily cohorts overlap. The empirical spread is wide. Analysis of pressurised water reactor programmes puts typical durations at roughly 60 to 80 months in standardised programmes, with the envelope covering around 95 per cent of observed projects running from about 50 to 250 months. We therefore model three durations rather than one: six years as a best-in-class case with mature technology and serial build, eight years as the central-duration assumption, and ten years as a conservative case capturing first-of-a-kind risk and programmes without standardisation or continuity.
Unit costs are expressed as overnight capital cost per GWe, with three bounding cases at US$6 bn, US$9 bn and US$12 bn per GWe. These are grounded in the IEA and NEA Projected Costs of Generating Electricity framework and in Lazard’s Levelized Cost of Energy+, which is explicit about how little public and observable data exists for new-build nuclear. Overnight capex is deliberate framing: it isolates the construction programme requirement. Outturn cost can be materially higher once cost of capital during construction, schedule extension, commodity and labour volatility and risk premia are added, which is precisely why schedule performance is a financial lever and not just a delivery one.
The headline results
Three benchmark cases carry the analysis, selected from the nine duration and cost combinations as stress tests of annual funding: a lowest-peak case of ten-year duration at low cost, a central case of eight years at base cost, and a highest-peak case of six years at high cost. The labels describe the annual funding stress each case creates, which is why the six-year duration, the most efficient delivery environment in the model, appears in the case with the highest peak.
Mean annual nuclear construction spend ranges from US$207 bn to US$491 bn a year. Peak annual requirements range from US$351 bn to US$854 bn, concentrated in 2040 and 2041. Cumulative capital commitment over 2026 to 2050 runs from about US$5.2 tn to US$12.3 tn.
In the model, shorter planned construction periods concentrate each cohort’s expenditure into fewer years, whereas longer periods spread spending across more years and require earlier starts. At a fixed unit cost and connection pathway, duration changes the timing and peak of expenditure, and the spread between the three benchmark cases also reflects their different unit costs of US$6 bn, US$9 bn and US$12 bn per GWe. These planned-duration effects should be distinguished from unplanned delay, which can increase financing and escalation costs.
Measured against what the world spends today
The comparison draws on the IEA’s World Energy Investment 2025 on a 2025 basis and uses three baselines:
- Nuclear investment, at about US$74 bn a year.
- Clean generation investment, renewable power and nuclear together, at about US$854 bn.
- A wider supply-side envelope of about US$2,002 bn, formed by adding the roughly US$1.1 tn the IEA records for oil, gas and coal supply to that clean generation figure.
The third baseline is deliberately broad: it measures the pool of supply-side capital that new nuclear competes with, and it is larger than investment in power generation alone, which the IEA puts at around US$1 tn a year across all technologies. The IEA’s headline clean energy total of about US$2.2 tn also includes grids, storage, efficiency and electrification, so it is a poor comparator for power station construction.
Mean annual nuclear spend comes out at between 280 and 664 per cent of today’s nuclear investment, and peak annual spend at between 474 and 1,154 per cent. Measured against clean generation, mean annual spend is 24 to 58 per cent of today’s level, and measured against the wider supply-side envelope it is 10 to 25 per cent. In the highest-peak case, peak nuclear construction spend approaches the whole of current investment in clean generation and about 43 per cent of the supply-side envelope, which is close to 85 per cent of the roughly US$1 tn the IEA records for all power generation investment.
Cumulatively, even the highest-peak case remains a minority share of the supply-side envelope if current levels were sustained over the same twenty-five years, although that comparison alone does not establish affordability, financeability or the availability of capital for individual projects.
Peak concentration is the binding constraint
What breaks the pathway is not the total. It is whether the system can absorb very large peak-year flows without triggering the failure modes the sector already knows well: supply chain bottlenecks, competition for scarce specialist labour, schedule stretch, rising contingency and a cost of capital that climbs as delay compounds. A programme can look entirely manageable in cumulative terms and still fail because its peak years demand a step-change that licensing capacity, heavy manufacturing and risk capital cannot deliver at the same moment.
There is also a displacement question that rarely gets asked plainly. If the overall supply-side investment envelope stays roughly where it is, a nuclear programme at this scale has to be funded by supplanting other supply-side investment. The only category large enough to absorb that is fossil fuel supply investment. It is not credible to assume renewables investment falls to make room, given electrification, security of supply and system resilience pressures all point the other way. Delivering the nuclear ambition alongside continued renewables expansion therefore requires the envelope to grow materially, fossil supply investment to contract materially, or both.
What this means for delivery and supply chain
The modelling has three direct implications for how the programme is run. The first is that delivery performance works as a system-level investment lever, because unplanned overruns differ from planned longer durations: a project that overruns keeps its capital committed while later cohorts start on schedule, so overlapping spend accumulates and financing and escalation costs rise. The second is that timing and sequencing matter as much as unit cost, since earlier starts and smoother profiles leave the build requirement unchanged while materially reducing peak stress. The third is that the implied build rate depends on a capability pipeline that has to be developed deliberately, because nuclear-grade forgings, certified component manufacturers, quality assurance capability, licensed construction labour and experienced regulators all have lead times measured in years, and in some cases a decade or more.
Supply chain readiness is usually where the delivery rate is actually set. Heavy manufacturing capacity and long-cycle components scale on the back of stable order books, not announcements. Without early visibility of demand and programme continuity, attempts to accelerate tend to raise cost and stretch duration, which pushes outcomes towards the conservative pathway rather than away from it.
The modelling is technology-neutral and does not separate large reactors from small modular ones. In practice the next ten to fifteen years of delivery will be dominated by large plants, given their maturity, licensing status and existing supply chains, with SMRs contributing progressively as designs, licensing pathways and delivery models mature. Accelerating SMR maturity and embedding deployment in repeatable delivery frameworks shortens durations, reduces unit costs and compresses learning curves, which moves realised outcomes towards the more optimistic pathways.
The judgement
The 2050 ambition is achievable, but only under a delivery model meaningfully different from the one that has characterised many recent nuclear projects. Programmes that behave like a series of bespoke megaprojects produce sharp peaks, volatile financing requirements and elevated risk premia. Programmes built on repeatable designs, serial build, earlier starts and disciplined sequencing reduce peak stress without changing the scale of the task.
Treated as procurement plus industrial strategy, the investment requirement becomes something markets and supply chains can organise around. Treated as a collection of disconnected megaprojects, the pathway becomes self-defeating: peaks become politically and financially intolerable, cost of capital rises, delays compound, and the ambition is eroded by the delivery dynamics it depends on.
The full report, including annual spend profiles for all nine duration and cost combinations, is available from NCL on request. If you are assessing programme exposure, supply chain investment cases or financing structures against this trajectory, get in touch.