AI Will Force a Nuclear Reckoning

Data centres, electrification and energy security are converging on a simple reality: intermittent power cannot carry a modern civilisation—and nuclear is the only scalable alternative.

Nuclear power is the closest thing humanity has to “always‑on” electricity. Across the world’s reactor fleet, plants routinely run at capacity factors around 80–95%, year after year, humming in the background while other sources ramp up and down with the weather or the gas price. A single large nuclear station can power millions of homes from a relatively compact site, while delivering low‑carbon electricity with a with a fuel supply measured in a few tonnes of uranium per year, rather than the hundreds of thousands of tonnes of coal, gas, that need to be extracted refined transported and burned annually, without the sprawling land area or millions of hours of sunshine or wind that renewables must harvest to match a single plant. In terms of reliability, land footprint, carbon footprint and sheer continuous output, as well as ongoing operating costs, nothing matches what a well‑run nuclear plant can do.

This matters now more than at any point since the first reactors were connected to the grid. Three great waves of demand are crashing into our electricity systems at the same time. In the developing world, hundreds of millions of people are still hungry for the kind of reliable power that unlocks factories, cold chains, hospitals, and real social mobility. In richer countries, the quiet, invisible growth of data—cloud computing, streaming, AI training and inference—is turning into a roaring appetite for power, with data centre demand expected to soar over the coming decade. Layered on top of this is the energy transition itself: the push to electrify vehicles, heat, and industry, while also replacing fossil‑fuel power plants with low‑carbon alternatives. The world is asking for more electricity, cleaner electricity, and more reliable electricity, all at once.

In that context, nuclear is back under serious consideration, not as a nostalgic “renaissance,” but as a hard‑headed recognition that there are very few technologies capable of delivering dense, round‑the‑clock, low‑carbon power at massive scale. Yet in the West, nuclear power faces two brutal, self‑inflicted obstacles: it has become astonishingly expensive to build, and painfully slow. Flagship projects have dragged on for decades and blown through budgets by tens of billions of dollars, making nuclear the poster child for cost and delay in. These are not small hurdles; they are the reason many planners reach for gas, wind, and solar first, even when they know those alone cannot carry the whole load.

But it was not always like this, and it is not like this everywhere. In the 1960s and 1970s, Western countries built entire nuclear fleets in a fraction of the time now considered “normal,” and at costs that, adjusted for inflation, look almost unbelievable. Even now, countries such as South Korea, China, and Russia continue to deliver new reactors at far lower costs and in far shorter timelines than Western projects. Somewhere along the way, the West took a turn, that transformed nuclear from a symbol of modernity into a byword for delay and overspend.

The Story of Nuclear Power

Nuclear power began with a weapon. The same discoveries in nuclear fission that produced the Manhattan Project and the atomic bombs over Hiroshima and Nagasaki also revealed an almost unbelievable density of energy locked inside the atom. By the end of the Second World War, governments and scientists were already asking whether this terrifying force could be tamed—first for submarines and ships, and then for something even more ambitious: turning the heat of controlled fission into electricity.

The shift from bomb to grid happened fast. In 1951, an experimental reactor in Idaho produced the first usable nuclear-generated electricity; a few years later, Eisenhower’s “Atoms for Peace” speech explicitly called for turning nuclear technology toward civilian power. Through the late 1950s and 1960s, the United States, the United Kingdom, France, Canada, and the Soviet Union all began to build commercial reactors, moving from prototypes to full‑scale power plants feeding national grids.

From the late 1960s into the 1980s, Western nuclear build‑out went into overdrive. In the early boom years, OECD countries built nuclear quickly and at costs that look almost unreal today. Between about 1970 and 1985, global nuclear capacity grew at roughly 15–17% a year, and most of that expansion was in the US, Western Europe and Japan. Typical construction times for these first‑ and second‑generation plants were on the order of 5–7 years from first concrete to grid connection, and overnight capital costs in many OECD projects were often in the range of roughly 1,000–2,000 dollars per kilowatt (in today’s money) as utilities standardised designs and rode down the learning curve. The result is that, even now, the bulk of nuclear electricity in the West still comes from reactors built in that 15‑year sprint, when capacity was doubling every few years.

Two major trends turned the nuclear energy renaissance on its head. First, the anti‑nuclear campaign model got industrialised. From the 1970s onward, Greenpeace and allied NGOs turned three quite different events – Three Mile Island, Chernobyl and later Fukushima – into a single, looming civilisational threat, folding bombs, waste and civil power into one undifferentiated fear. The imagery was powerful; the reality far less so. Three Mile Island killed no one; Chernobyl a first generation reactor had the only ever major nuclear failure , UN and UNSCEAR assessments put confirmed immediate deaths in the dozens and project total long‑term cancer deaths in the low thousands, not the hundreds of thousands implied in many narratives. Even Fukushima’s direct radiation health effects are expected to be so small they will likely never show up clearly in population statistics. Yet fear, stigma and political backlash did more lasting damage than the physics of the accidents themselves.

Second, this fear‑driven politics collided with financialization and de‑industrialisation in the West. Nuclear stopped being treated as strategic infrastructure, heavy industry and engineering capability were offshored as “dirty” or “non‑core”. Each new plant proposal became a perfect storm: a technically scrutinised gauntlet, and attacked by professionalised campaigns designed to maximise public anxiety.

After 1985 across the OECD, total nuclear capacity barely grew at all over the next three and a half decades: from the mid‑1980s to around 2020, net capacity in OECD countries increased by only a few tens of gigawatts, amounting to something like a 10–20% cumulative gain on the fleet built in the original boom. At the same time, the economics and timelines of new builds deteriorated sharply. Average construction times for large Western projects stretched towards 10–15 years, and estimated overnight costs climbed from under 2,000 dollars per kilowatt in the late 1990s to around 3,800–5,000 dollars per kilowatt or more in many OECD studies by the late 2000s and 2010s, with some flagship projects far exceeding that. The latest total cost estimates for Hinkley Point C are now in the region of £35–46 billion for about 3.2–3.3 GW of capacity. That implies a capital cost on the order of £10,000–14,000 per kW (roughly $12,000–17,000 per kW). While this is the worst of the time and cost inflation in the world, it serves as a cautionary tale to utilities globally when they consider putting in new nuclear capacity

China’s nuclear programme tells the mirror‑image story. Starting from essentially zero commercial capacity in 1990, China had only a few gigawatts installed by the early 2000s, but then began a sustained build‑out: by the early 2020s its operating capacity had passed 50 GW, with official plans suggesting around 70 GW by 2025 and much more beyond. That implies a double‑digit annual growth rate in nuclear capacity over the past two decades, with some analyses projecting around 7% per year expansion continuing for at least another 10–15 years as new reactors move through China’s five‑year plans. Most Chinese units have been completed in roughly 5–7 years on average, and published estimates and benchmarking exercises suggest overnight costs often in the low‑to‑mid 2,000 dollars per kilowatt range—roughly half or less of many recent Western projects—thanks to standardised designs, domestic supply chains and a single, deliberate national programme. In other words, while the OECD has spent 35 years adding only a thin layer of capacity on top of its legacy fleet, China has built an entire nuclear system from scratch, at growth rates and unit costs that look much closer to the West’s own experience in the 1970s than to its present‑day reality.

The technology that built the original Western nuclear renaissance was the Westinghouse‑style pressurised water reactor, born from the US Navy submarine programme and then standardised for civilian grids from the late 1950s onward. By the 1970s, PWRs had become a global template: hundreds of units were built on variations of this design in the US, France, Japan and elsewhere, with utilities and vendors learning how to repeat the same basic plant again and again. That standardisation—same core design, same major components, same construction sequence—was exactly what drove down build times and kept early capital costs relatively low compared with what came later.

China’s modern nuclear programme is, in a sense, a direct heir to that Western work. When Beijing decided to scale nuclear in the 2000s, it deliberately imported and licensed Western designs like the Westinghouse AP1000 as its “technology basis,” then used extensive technology‑transfer agreements to localise manufacture and evolve Chinese variants such as the CAP1400. Only after building up this fleet of relatively conventional PWRs at scale has China begun to lean into more experimental options, including its molten‑salt thorium test reactor (TMSR‑LF1) in Gansu, which has already achieved thorium‑uranium fuel conversion and first criticality. In other words, China first copied and industrialised the Western standard, and only then started optimising and experimenting at the edges.

In a world where China can now take a Westinghouse‑derived design and deliver a standardised reactor in 5–7 years, the obvious question is: if the technology is familiar and fundamentally standardised, why has the West struggled so badly to build Hinkley‑style projects using the same technology, on time and anywhere near budget, when the 1970s plans assumed this would only get easier?

How Big things get Done

Bent Flyvbjerg is an economic geographer and project‑management scholar widely described as the world’s leading expert on megaprojects. His book How Big Things Get Done, is his accessible synthesis of that research on over 16000 global megaprojects from dams and railways to nuclear reactos (I cannot recommend it more). The book’s starting point is what he calls the “iron law of megaprojects”: most big projects come in over budget.

Flyvbjerg’s work is really about a civilisational skill that the West has lost: the ability to deliver big, complex projects repeatably, on time and on budget. He shows that most megaprojects in rich democracies fail in exactly the same way – bespoke designs, one‑off consortia, fragmented decision‑making, and politics layered onto every technical choice – while the rare successes come from doing the opposite: ruthless standardisation, lots of repetition, and treating projects as an industrial product rather than a heroic one‑off. Nuclear in the West sits right in the crosshairs of that diagnosis. The book is ultimately a critique of how rich democracies now do infrastructure, and a playbook for how to recover the ability to deliver big things reliably.

China’s nuclear programme is like a live demonstration of Flyvbjerg’s playbook. It imported standard Western reactor designs, licensed them, and then did what the West stopped doing: building them over and over in similar configurations, with centralised planning, long‑term state backing and little tolerance for fear‑based campaigning. Public, adversarial protest is tightly constrained, but that has not meant lower safety; Chinese regulators and designers have systematically incorporated lessons from Three Mile Island, Chernobyl and Fukushima, adding layers of passive safety and severe‑accident management much as Western engineers do. That is the rational response to accidents: treat them as information, improve the design, tighten procedures, and then build the next unit slightly better. In contrast, many Western countries drifted into treating each reactor as a semi‑unique project, with bespoke design changes, site‑specific political battles and years of hearings – a permanent “start from scratch and argue about everything” mode. Hinkley Point C is the emblem of this pattern: an evolutionary PWR design that should benefit from decades of learning, yet is delivered as a one‑off megaproject mired in delay, cost escalation and process. The tragedy is not only the wasted money and time, but that a system which once knew how to build big, safe reactors as a matter of routine has allowed itself to be paralysed by fear, proceduralism and institutional habits that almost guarantee failure. A minimum of 60% of the costs of Western nuclear delivery is managing read tape. Industry cost breakdowns show less than 40% of the upfront cost of a new reactor are labour and raw materials and equipment; the rest sits in design and engineering, licensing, project management where decades of regulatory accretion, proceduralism and low productivity have piled up.

Nuclear power is the only technology that even comes close to meeting the triple challenge now bearing down on the world: a deepening energy crisis, the global push to electrify everything, and the need to decarbonise without accepting permanent energy austerity. Nothing else offers the same combination of 24/7 reliability, tiny land footprint, astonishing fuel density and multi‑decade lifetimes. A single large plant can quietly deliver gigawatts of clean power for 60–80 years, day and night, in all weather, with a fuel supply measured in tonnes per year rather than endless trains of fossil fuels or vast landscapes covered with panels and turbines. And yet, in the West, the industry that could have underpinned such a future was systematically strangled by a fear campaign and red tape.

The Western nuclear legacy now lives most vividly in competitor nations. While Western countries are mired in stagnating productivity, ballooning megaproject costs, and political systems that can barely approve a transmission line, their former technological advantages are being quietly scaled elsewhere. The irony is brutal: the very reactors that underpin China’s nuclear programme descend directly from Western designs, refined by Western engineers, under Western safety regimes, while Western governments struggle to complete even a handful of new units.

The Second Nuclear Rennaissance

A distinctly capitalist force is beginning to pull nuclear back into the frame: the rise of AI and hyperscale computing. Modern data centres – especially those built for AI – are voracious electricity machines. They cannot live on intermittent power, and they are running into pushback from local grids and communities that do not want huge new loads tacked onto already strained systems. The logical response is to bring dedicated generation to the compute: firm, local, controllable power that can run flat‑out for decades. That points directly at nuclear. Unlike many governments, the large technology firms know how to deliver massive, complex projects and ruthlessly optimise cost, schedule and design when their business depends on it.

Defence and security actors have parallel needs. Militaries require sovereign, resilient power for bases, ports, radar, fuel synthesis and space assets – power that cannot be switched off by a pipeline disruption, a cyberattack on the grid, or a geopolitical embargo. Small, robust reactors that can be built in factories, shipped as repeatable units and operated safely in demanding environments are a natural fit. The reality is the first reactors were built within the defence industry, and nuclear submarines are the only place where nuclear skill sets have survived in the West. As early projects prove themselves on bases and in industrial clusters, they will create reference cases, skills and supply chains that civilian buyers can tap into. The first new military‑hosted reactors in the United States are now approved and more projects are in the pipeline.

If AI hyperscalers and defence planners converge on the same solution – distributed but standardised nuclear units, delivered at industrial scale – a genuine nuclear resurgence becomes possible, not relying on politicised government utility scale projects, but as a hard‑headed response to energy scarcity and and the needs of industry. In that ecosystem, nuclear is not a fragile, over‑promised “renaissance” but a workhorse of an energy‑rich civilisation: fleets of advanced and modular reactors quietly powering data centres, industrial hubs, desalination plants, green fuels and cities, stitched into renewables‑heavy grids. The West would not just be buying reactors from abroad; it would again be designing, building and exporting the core machinery of a prosperous, decarbonised world. That is the hopeful, visionary path: not a return to the 1970s, but a fusion of digital and physical, where the same ingenuity that built the information age finally turns, in earnest, to rebuilding the energy systems that make it possible.

If that happens, the next chapter of nuclear will not be written by the same slow, politicised institutions that suffocated the last one. It will be driven by innovators, technologists, capital allocators and security communities that cannot afford to wait 15 years for each plant and do not have the luxury of indulging in performative paralysis. Out of the ashes of an industry the West once led and then nearly lost, a very different nuclear ecosystem could rise: modular, standardised, globally scalable, anchored by the most energy‑hungry sectors of the digital and defence economies. For a world that wants to stay rich, go green and keep the lights on, that phoenix may be the only realistic route to an energy‑abundant, decarbonised future.

If you have made it this far thanks you, please consider subscribing (its free) for clear, grounded analysis on the systems that actually power the world—and share if you think we need a more honest conversation about energy.