The making and unmaking of nuclear energy

The electricity system is being asked to do several hard things at once: decarbonise, stay reliable, electrify industry and transport, and absorb new load from data centres and AI computation. After two decades in which nuclear's share of global generation drifted down from a late-1990s peak near 17.5 percent to roughly 9 percent today, the technology has re-entered strategic debate. At COP28 in December 2023, more than twenty governments pledged to triple installed nuclear capacity by 2050—a coalition that grew past thirty states through 2024 and 2025—while the United States, the United Kingdom, France, and Japan reframed nuclear as central to energy security after Russia's invasion of Ukraine exposed the cost of fossil dependence (IEA 2025).

The debate, however, is too often flattened into ideology: one camp treats nuclear as the indispensable answer to climate and security, another as too slow, too costly, and too politically brittle to matter. The more useful frame is institutional. Nuclear succeeds when four pillars reinforce one another—engineering capability that lets designs be standardised and repeated; patient capital that can survive decade-long construction; regulatory continuity that keeps safety credible without making each plant a one-off; and public legitimacy that holds siting and licensing together over a sixty-year asset life. When all four align, nuclear delivers cheap, durable, firm power. When any one breaks, projects stall, costs compound, and confidence collapses.

Civilian nuclear power was born from military programmes and matured under state sponsorship. The United States' Atoms for Peace initiative (1953) and the first grid-connected reactors of the late 1950s established a template in which vertically integrated utilities and sovereign balance sheets absorbed early risk, while public narratives framed the atom as abundance, modernity, and strategic autonomy.

France is the canonical success. After the 1973 oil shock, the Messmer Plan committed the country to a standardised fleet built by a single architect-engineer pairing on a handful of repeated designs. Within roughly fifteen years France constructed fifty-six reactors and reached about 70 percent nuclear electricity—the fastest decarbonisation of a major grid in history. The decisive variable was repetition: the same designs, the same teams, predictable financing, and a regulator working with familiar technology (Lovering, Yip, and Nordhaus 2016).

South Korea followed a comparable programmatic path with its APR-1400 design and turned it into an export, completing the four-unit Barakah plant in the United Arab Emirates, whose final reactor connected to the grid in 2024. China is now executing the same playbook at greater scale, building reactors on standardised Hualong One and CAP1000 designs in roughly five to seven years each and running the world's largest construction pipeline. Where nuclear was embedded in a durable industrial system, learning compounded rather than resetting at every project.

From the late 1970s the model fractured across much of the West. Construction complexity, rising interest rates, and stretched timelines made plants acutely sensitive to delay and financing cost, while liberalised power markets introduced short-horizon capital that fit poorly with decade-long delivery risk.

Accidents reset public tolerance and the regulatory baseline. Three Mile Island (1979), Chernobyl (1986), and Fukushima (2011) each produced thicker regulation, slower licensing, and higher capital costs—and, in Germany's case, an outright exit, completed with the shutdown of its last reactors in April 2023. As order books emptied, supply chains atrophied: heavy forging capacity, specialised vendors, and skilled trades dispersed, so each new project effectively rebuilt a lost industry from scratch.

Recent flagship builds illustrate the cost of that lost capability. Vogtle Units 3 and 4 in Georgia, the first new US reactors in three decades, came online in 2023 and 2024 at roughly 35 billion dollars against an original budget near 14 billion and some seven years late. France's Flamanville 3 EPR connected to the grid in December 2024, about twelve years behind schedule and several times its initial cost, and Finland's Olkiluoto 3 told a similar first-of-a-kind story. Each delay amplified both expense and distrust, creating a self-reinforcing cycle in which poor performance became the expectation rather than the exception.

Three forces revived the case. First, decarbonisation raised the value of firm, dispatchable low-carbon power in grids increasingly dominated by variable wind and solar. Second, the post-2022 energy-security shock made fuel-secure domestic baseload politically attractive. Third—and most novel—electricity demand is rising again after two decades of flat consumption in advanced economies, driven by electrification and, dramatically, by data centres and AI computation; the IEA projects global data-centre electricity use roughly doubling by 2030.

That last force pulled an unexpected actor into nuclear finance. In September 2024 Microsoft and Constellation agreed to restart the undamaged Unit 1 at Three Mile Island, rebranded the Crane Clean Energy Center, under a twenty-year power-purchase agreement; Amazon backed X-energy's small modular reactors and contracted output from Talen's Susquehanna plant; Google signed with Kairos Power; and Meta issued a request for proposals for up to four gigawatts of nuclear capacity. For the first time, corporate balance sheets and long offtake contracts are supplying the patient demand signal that nuclear finance has always lacked.

Small modular reactors are the other half of the revival thesis, promising factory fabrication, standardisation, smaller absolute capital, and easier siting. The reality is more sober. NuScale's flagship US project was cancelled in November 2023 as costs climbed, a reminder that smaller does not mean automatically cheaper. Yet TerraPower broke ground on its Natrium plant in Wyoming in 2024, GE Hitachi's BWRX-300 advanced toward construction at Ontario's Darlington site, and Rolls-Royce, Holtec, and X-energy pressed forward. SMRs may redistribute risk and improve repeatability, but they cannot abolish first-of-a-kind risk; their economics depend on building many identical units, which is precisely the programmatic discipline the West lost.

A nuclear revival is also a fuel-supply question, and here the geopolitics are stark. Uranium prices broke above 100 dollars per pound in early 2024 for the first time since 2007. Mined supply is concentrated—Kazakhstan's Kazatomprom alone provides around 40 percent of the world's uranium—but the more strategic chokepoint sits downstream in conversion and enrichment, where Russia controls roughly 40 percent of global enrichment capacity.

The war in Ukraine turned that dependence into a vulnerability. The United States enacted a ban on Russian enriched-uranium imports in 2024, with limited waivers through 2028, and moved to fund domestic conversion and enrichment. Advanced reactors compound the problem: many require high-assay low-enriched uranium (HALEU), enriched between 5 and 20 percent, for which Russia was until recently effectively the only commercial supplier. Building Western HALEU capacity is now both a strategic priority and a genuine bottleneck that could throttle the very SMR timeline the revival depends upon.

Reactor exports are themselves instruments of statecraft. To buy a reactor is to enter a sixty-year relationship spanning fuel, components, training, and financing. Russia's Rosatom and China's state nuclear firms have used export-credit-backed reactor diplomacy aggressively across emerging markets, while the United States, France, and South Korea compete to anchor allied standards. Nuclear's future is therefore entangled with industrial sovereignty as much as with climate arithmetic: countries revisiting the atom are also choosing whose technology they license, whose fuel they buy, and whose engineers they train.

The comparative record yields a clear conclusion: nuclear scales when projects become programmes. Standardised designs, repeat construction, stable regulation, patient state-backed finance, and a protected workforce compound into falling costs; bespoke megaprojects in stop-start political environments do the opposite. Small modular reactors are promising precisely because they lean into repetition, but they will validate the thesis only if order books are deep enough to capture the learning curve.

For decision-makers the implication is selective rather than universal. Countries with strong demand growth, credible state capacity, industrial financing tools, and—increasingly—corporate offtake can justify new nuclear as part of a clean-firm portfolio. Where those institutions are absent, renewables, storage, transmission, and demand flexibility will deliver faster near-term gains even as nuclear remains strategically attractive in principle. Nuclear is neither an obsolete relic nor an automatic climate solution; it is a high-discipline infrastructure system whose return depends on rebuilding the institutions that made it work the first time.