Nuclear Energy: The Promise, the Accidents, and the Comeback

Follows: How a Nuclear Reactor Works (Physics 14) and Energy: The One Currency of the Universe (Physics 2)

In March 1979, the reactor known as Unit 2 at Three Mile Island in Pennsylvania partly melted down, and the American nuclear industry never fully recovered. Its twin, Unit 1, stood a few hundred metres away and kept running for another forty years. It shut in 2019 because it couldn't pay its way. Then, in September 2024, its owner signed a 20-year deal with Microsoft to bring it back to life, to feed the electricity demand of computing. It now has a new name and a target date of 2027.

That one site holds the whole story: the promise, the accident, the long decline and the unlikely comeback. This deep dive tells that story from the first chain reaction in 1942 to the state of things at the end of October 2026. It also sets out why intelligent people still disagree so strongly about it. The physics (what fission is, why a reactor can be steered, how Chernobyl's design failed) is in How a Nuclear Reactor Works; this piece is about history, money, risk and politics.

In this deep dive

1. From a chain reaction to Atoms for Peace (1942–1957)
2. The boom, the backlash, and Three Mile Island
3. Chernobyl, Fukushima, and how to count the dead
4. France and Germany: two countries, two answers
5. Waste: the problem that has to outlast us
6. The comeback: nuclear power in 2026
7. Fusion: the other nuclear
8. The strongest case for, and against

Part 1 — From a chain reaction to Atoms for Peace (1942–1957)

The idea came before the machine, and it came as a fear. Radioactivity, discovered in the 1890s, showed that atoms hold energy on a scale nothing in chemistry could match (see Marie Curie: Two Nobel Prizes, One Radical Idea). For forty years nobody knew how to release it on purpose. Then, in December 1938, Otto Hahn and Fritz Strassmann found that uranium hit by neutrons split in two, and within weeks Lise Meitner and Otto Frisch had explained it. Physicists saw at once that the split also released neutrons, which could set off a chain reaction. With war coming, the first use anyone imagined was a weapon.

The first chain reaction was achieved quietly. On 2 December 1942, under the west stands of Stagg Field, the University of Chicago's disused football stadium, Enrico Fermi's team pulled the last control rod out of a pile of about 45,000 graphite blocks studded with uranium. At 3:25 in the afternoon, the neutron count kept rising on its own. Chicago Pile-1 was critical. It produced about half a watt, too little to boil a kettle in a week, and the first run lasted only minutes. Forty-nine people watched. Eugene Wigner had brought a bottle of Chianti, which they drank from paper cups and signed.

"The Italian navigator has landed in the New World." Arthur Compton's coded phone report to James Conant that the chain reaction had worked.

Chicago Pile-1 was a step in the Manhattan Project, and the project is the original sin that nuclear power has never fully shaken off. At its peak it employed about 130,000 people and had spent about $2.2 billion by the end of the war. It built secret towns: Oak Ridge in Tennessee to enrich uranium, Hanford in Washington to make plutonium, Los Alamos in New Mexico to design the bombs. The first test, Trinity, was on 16 July 1945. On 6 and 9 August, bombs destroyed Hiroshima and Nagasaki. Estimates of the deaths they caused range from about 150,000 to 246,000.

Two legacies of these years shaped everything after. Commercial reactors grew up inside military programmes built in secret and at any cost. And the gap between power and weapons is narrower than people would like. A reactor runs on uranium enriched to a few per cent uranium-235, a bomb needs it very highly enriched or needs plutonium, and the plants that make one can, with intent, make the other.

On 8 December 1953, President Dwight Eisenhower stood before the United Nations General Assembly in New York and tried to change the subject. With the superpowers racing to build hydrogen bombs, Eisenhower spoke of "the fearful atomic dilemma", and proposed that nuclear powers pool materials under an international agency for peaceful uses, above all electricity. His hope was that "the miraculous inventiveness of man shall not be dedicated to his death, but consecrated to his life." The speech led to the founding of the International Atomic Energy Agency and to a programme that shared reactor know-how and fuel with friendly countries.

It is still argued over. Supporters see the moment nuclear technology turned toward medicine, research and power. Critics point out that the United States went on exporting highly enriched uranium for research reactors (over 25 tons to 30 countries), now regarded as a proliferation risk. They note that Israel's and Pakistan's first reactors came out of the programme, and that it gave political cover while the American arsenal grew from about a thousand weapons to about twenty thousand during Eisenhower's presidency. Both readings are true at once.

Who was "first" with nuclear electricity depends on the definition.

Shippingport shows how much of the industry came from the navy. Rickover's programme had developed the pressurised-water reactor to drive submarines. Its prototype started in Idaho in March 1953, and USS Nautilus was launched in 1954. That compact, water-cooled design was adapted for land, and it is still by far the most common type: about 316 of the world's roughly 440 operable reactors are pressurised-water reactors. The economist Robin Cowan argued in 1990 that this was a case of "technological lock-in": the reactor that won was the one the navy had already paid to develop, not necessarily the best design for a power grid.

Part 2 — The boom, the backlash, and Three Mile Island

For about twenty years, nuclear power grew faster than almost any energy technology before it. Westinghouse's Yankee Rowe plant (250 MW) started in 1960. By the end of the decade, utilities were ordering reactors of more than 1,000 MW, as big as the largest coal plants. World nuclear generation rose from about 79 terawatt-hours (TWh) in 1970 to about 2,000 TWh in 1990, a twenty-five-fold increase in twenty years. Its share of the world's electricity peaked at about 17.4% in 1996.

The mood of the boom had been set by one phrase. In September 1954, Lewis Strauss, chairman of the US Atomic Energy Commission, told science writers that "it is not too much to expect that our children will enjoy in their homes electrical energy too cheap to meter". The line became the industry's most quoted embarrassment. His son later claimed he had been thinking of fusion, then secret. A historian at the Nuclear Regulatory Commission, Thomas Wellock, found no evidence of that in Strauss's papers, and the rest of the speech was about fission. Either way, the promise set an expectation the industry never met.

In twenty years, nuclear generation grew twenty-five-fold. In the thirty years after that, its share of the world's electricity halved.

The oil shocks of the 1970s seemed to guarantee the boom. Oil-importing countries, France most dramatically, turned to reactors for energy security. But in the United States the boom was already faltering before the famous accident. Reactors were getting bigger, each one closer to a custom project, and costs and construction times grew with them. Many reactor orders from the 1970s were cancelled.

At the same time, a popular movement against nuclear power took shape. In West Germany, protests against a planned plant at Wyhl in the 1970s turned public opinion. The movement fed into what became the Green Party. Its objections were not only about accidents. They were about waste, about the link to weapons, and about a technology so complex and centralised that only states and giant firms could run it. In March 1979 Hollywood released The China Syndrome, a thriller about a cover-up at a reactor. Twelve days later, life imitated it.

Three Mile Island, 28 March 1979. At about 4 a.m., a fault stopped the main feedwater pumps of Unit 2. A relief valve opened as designed but then stuck open, and a light in the control room told the operators it had closed. Coolant kept escaping while the staff, misled by their instruments and lacking any reliable way to tell whether the core was covered with water, took actions that left it uncovered. Nearly half of the core melted.

The releases of radioactivity were very small. The US Nuclear Regulatory Commission's figure is that the roughly two million people around the plant received on average about 1 millirem above background. A chest X-ray is about 6 millirem, and the area's natural background is about 100 to 125 millirem a year. More than a dozen studies have found no detectable health effects. The panic was real, though. The governor advised pregnant women and preschool children within five miles to leave, and about 140,000 people left the area within days, though a survey that April found 98% of them back home within three weeks. The clean-up took until 1993 and cost about $1 billion.

At Three Mile Island no one died. The industry nearly did.

The accident changed regulation more than health. The NRC put resident inspectors at every plant, overhauled operator training and control-room design, and created a 24-hour operations centre. The industry set up its own body to share lessons and police itself.

Part 3 — Chernobyl, Fukushima, and how to count the dead

Chernobyl, 26 April 1986. At unit 4 of the Chernobyl plant in Soviet Ukraine, a badly run safety test on a reactor design with dangerous built-in flaws (explained in How a Nuclear Reactor Works) caused a power surge and an explosion that blew the reactor open, sending radioactive material across much of Europe. Two workers died that night. Of the 134 people confirmed with acute radiation sickness, 28 died within weeks. Nineteen more of them died between 1987 and 2004, though not necessarily from radiation.

About 116,000 people were evacuated by mid-May 1986, and 220,000 more were resettled later. The exclusion zone grew from a 30-kilometre circle to about 4,300 square kilometres. The most clearly established health effect came from radioactive iodine in milk: children drank it, and thousands of them later developed thyroid cancer. UNSCEAR, the UN's scientific committee on radiation, put the number at more than 6,000 cases by 2005, of which 15 had been fatal. Thyroid cancer is usually treatable, which is why the deaths are few next to the cases.

A steel arch, the New Safe Confinement, was slid over the ruined reactor in November 2016. On 14 February 2025 a drone struck it and breached its outer cladding. A second Ukrainian nuclear plant, Zaporizhzhia, has been under Russian occupation since March 2022, with all six reactors held in cold shutdown since that September.

Fukushima Daiichi, 11 March 2011. At 2:46 p.m. a magnitude 9.0 earthquake struck off Japan's north-east coast. The reactors running at Fukushima Daiichi shut down automatically, as designed. The tsunami that followed reached about 15 metres at the site and flooded the backup generators. With no power to remove their decay heat, the cores of units 1 to 3 largely melted within three days. Hydrogen explosions wrecked the reactor buildings. The accident was rated at level 7 on the international scale, the same as Chernobyl.

The earthquake and tsunami themselves killed nearly 20,000 people, with more than 2,500 missing, by official counts. Radiation sickness killed no one, though in 2018 the government accepted that one plant worker's death from lung cancer was linked to his exposure. The evacuation is a different matter. Around 160,000 people left their homes, and by September 2020 Fukushima prefecture had recorded 2,313 "disaster-related deaths" among evacuees: people, about 90% of them aged 66 or over, who died of the stress, disruption and interrupted care of being moved. About 27,000 people were still displaced as of November 2025. In 2021 UNSCEAR concluded that "no adverse health effects among Fukushima residents have been documented", and that future effects from radiation were unlikely ever to be detectable.

The uncomfortable conclusion is that at Fukushima the evacuation seems to have caused far more deaths than the radiation it was meant to prevent. Some argue it was too large and too long. Others reply that officials could not know the doses in the chaos of March 2011, and that the evacuation was itself a cost of the accident.

How many did Chernobyl kill? Here the numbers diverge, and the reason is scientific as much as political.

Almost all the difference comes from one question: does a very small dose of radiation, spread over millions of people, cause a very small number of cancers? The linear no-threshold model says yes. Risk falls with dose but never reaches zero, so a tiny risk multiplied by a huge population gives thousands of deaths. Regulators use it because it is cautious. But at low doses the extra cancers are far too few to see against the normal background: cancer already causes nearly one death in six worldwide. UNSCEAR's position since 2012 has been that it does not recommend multiplying very low doses by large numbers of people to estimate deaths. The higher projections are not invented; they come from a model that can't be checked at those doses, in either direction.

Deaths per unit of energy. One way to compare energy sources is to count deaths from accidents and air pollution per terawatt-hour of electricity. A terawatt-hour is a year's electricity, counting homes, industry and everything else, for about 80,000 people in the United States or about 240,000 in Britain. Our World in Data's comparison, by Hannah Ritchie, uses the lower, UNSCEAR-based Chernobyl figure (433 deaths) and the Fukushima evacuation deaths (2,314):

What if the higher Chernobyl estimates are right? The calculation is easy to redo. Nuclear plants generated about 96,900 TWh between 1965 and 2021. Using the Chernobyl Forum's 4,000, nuclear comes out at about 0.07 deaths per TWh. Using Greenpeace's 93,000, about 1.0, which is still a third of gas and a twenty-fifth of coal. Only the most extreme estimate, nearly a million, would put nuclear above gas. On almost any mainstream number, electricity's main killer is not radiation but smoke from burning coal, oil and wood, which kills quietly every day rather than spectacularly once a generation.

Part 4 — France and Germany: two countries, two answers

Two rich, democratic neighbours made opposite choices, and each makes sense from inside its own history.

France went in. In the 1970s France had strong heavy engineering but few energy resources of its own. Most of its electricity was made by burning imported oil. A few months after the 1973 oil shock, on 6 March 1974, Prime Minister Pierre Messmer announced a crash programme of reactors using licensed Westinghouse technology. A popular slogan of the period ran: "In France, we do not have oil, but we have ideas." About 56 reactors went into service over the following fifteen years. By 1990 nuclear supplied three-quarters of France's electricity, and close to 80% by the mid-2000s.

The defenders' case is in the results. In 2025, France's 57 operating reactors produced about 373 of its 547 TWh, around 68%. Its electricity emitted about 41 grams of CO₂ per kilowatt-hour, less than a tenth of the world average. The programme also built a national industry and a skilled workforce.

The critics' case is also real. When one technology supplies most of a country's power, a fleet-wide problem is a national one. In 2022, stress corrosion found in reactor pipework cut nuclear output to 282 TWh against a ten-year average of 395, during a European energy crisis. And France lost the knack of building. Flamanville 3, its first new reactor in years, began construction in December 2007 and was meant to start in 2012. It reached the grid on 21 December 2024. EDF's own cost estimate rose from €3.3 billion to €13.2 billion, and France's national audit court, the Cour des comptes, put the full cost, including financing, at up to €19.1 billion in 2020. A 2014 law to cut nuclear's share to 50% was abandoned in 2023. In February 2022 President Macron announced six new large reactors (EPR2s), with eight more under consideration. EDF's estimate for the first six rose to €72.8 billion in December 2025.

France proves that a country can decarbonise its electricity with nuclear power. Flamanville proves how hard it is to do it again.

Germany got out. West Germany's anti-nuclear movement was rooted in the Wyhl protests and in long fights over a proposed waste site at Gorleben. After Chernobyl, the Social Democrats resolved to leave nuclear power. In 2000–2001 a Social Democrat–Green government agreed a phase-out with the utilities, limiting each reactor's lifetime output. In 2010 Angela Merkel's government extended the reactors' lives. Then came Fukushima. In March 2011, eight older reactors were shut immediately, and at the end of June the Bundestag voted 513 to 79 to close all of them by 2022. After Russia cut gas supplies, the last three ran a few extra months and closed on 15 April 2023. In 2021 the state agreed to pay the utilities €2.4 billion in compensation.

Germany's case, in its defenders' terms, was never only about fear. Its supporters argued that no country had yet opened a permanent repository for waste. They argued that a densely populated country should not accept even a small chance of a disaster like Fukushima's, and that money and political attention should go to renewables, which they expected to become cheaper and safer. On that last point they have a record to point to: renewables rose from about 17% of German electricity in 2010 to about 59% in 2025, and coal fell from 42% to about 21%.

Critics answer that the order of events was wrong. In 2010 nuclear supplied about 22.5% of Germany's power with almost no carbon. Had the reactors stayed open, the new renewables could have pushed out coal far faster. A study by the economists Stephen Jarvis, Olivier Deschenes and Akshaya Jha estimated that the lost nuclear output was replaced mainly by coal and imports, at a social cost of about $12 billion a year, more than 70% of it from deaths caused by air pollution. Germany's electricity in 2025 emitted about 330 grams of CO₂ per kilowatt-hour, eight times France's. In May 2025, under Chancellor Friedrich Merz, Germany dropped its opposition to treating nuclear power on a par with renewables in EU climate policy.

Neither side has to be foolish for this to make sense. France chose to accept nuclear risk to escape fossil fuels. Germany chose to escape nuclear risk and accepted burning coal for longer. They weighed the same risks differently.

Part 5 — Waste: the problem that has to outlast us

Every reactor produces spent fuel: intensely radioactive ceramic pellets, still in their metal tubes, that must be kept cool and shielded. The odd thing about nuclear waste is how little there is. The United States has accumulated about 90,000 tonnes of spent fuel from commercial reactors since the 1950s. Stacked together it would cover a single American football field to a depth of under ten yards. The country adds about 2,000 tonnes a year. After five years in a reactor, the fuel still holds more than 90% of its potential energy, which is why some argue it should be recycled rather than buried; the United States does not currently recycle it. Much of a coal plant's waste goes up the chimney and into everyone's lungs. Nuclear waste is small, solid and tracked, which is both why it can be managed and why it is so feared: it sits there, and someone has to look after it for a very long time.

Spent fuel spends at least five years cooling under water in pools at the reactor, then often goes into sealed casks shielded by steel and concrete. This works, but it is meant to be temporary. The permanent answer that is "widely agreed to be the best solution", in the World Nuclear Association's words, and is official policy in Finland, France, Sweden and the United States, is deep geological disposal: put the waste in stable rock, deep enough and with enough barriers that it stays isolated for longer than any human institution has lasted.

The United States tried and stalled. In 1987 Congress picked Yucca Mountain in Nevada as the only site to study. In 2002 it overrode Nevada's objection and approved the site. Then the Obama administration withdrew the licence application in 2010, and Congress stopped funding the project in 2011. The site sits unused, and American spent fuel sits where it was made.

Finland went ahead. At Olkiluoto, on the west coast, the company Posiva has dug a repository called Onkalo more than 400 metres down in the bedrock. The method, developed in Sweden, puts the fuel in a canister sealed inside copper, packs each one into its own hole with bentonite clay, and backfills the tunnels. Construction was licensed in 2015, full-scale testing began in August 2024, and on 22 October 2026 the Finnish government granted Posiva the world's first operating licence for the final disposal of spent nuclear fuel, valid to 2070. Disposal begins once commissioning tests are approved by the regulator; no start date has been announced. The licence covers 6,500 tonnes of spent fuel, and current plans have the repository sealed for good around 2120. Sweden's repository at Forsmark is expected in the 2030s, and France's, Cigéo, around 2050.

The open question is less technical than moral. A repository asks a society to make a decision that binds people thousands of generations away. The documentary Into Eternity (2010) is built around one problem: how do you warn people not to dig here, in a language no one will speak? Supporters answer that leaving the waste on the surface for ever is the worse burden to hand on. Critics answer that a technology which needs such a promise should not be expanded until the promise has been kept somewhere. Finland is now the test case.

Part 6 — The comeback: nuclear power in 2026

The numbers, first, as of the end of October 2026. The IAEA's reactor database (PRIS) lists 417 reactors in operation in 31 countries, with 380 gigawatts of capacity, plus 21 in "suspended operation" in two countries. It lists 78 under construction in 17 countries. The World Nuclear Association, which counts differently, gives 443 "operable" reactors and 79 under construction. Nuclear supplied about 2,600–2,800 TWh in 2025 (sources differ on the exact figure), about 9% of the world's electricity, roughly half its 1996 peak share.

Source: IAEA PRIS, read 31 October 2026.

China is building almost half of all the reactors under construction in the world. The West's recent record is harder to admire. America's only new reactors in a generation, Vogtle 3 and 4 in Georgia, entered service in July 2023 and April 2024, about seven years late. Their cost rose from an original estimate of about $14 billion to more than $30 billion, and Westinghouse went bankrupt along the way in 2017. Britain's Hinkley Point C is now estimated at £35 billion in 2015 prices, with its first unit not expected before about 2030.

So why the talk of a comeback? Three things changed. Climate policy made low-carbon power that runs day and night look valuable. Russia's invasion of Ukraine made energy security urgent again, as in the 1970s. And computing began to demand electricity on a new scale. A US government report from Lawrence Berkeley National Laboratory found that data centres used about 4.4% of American electricity in 2023 (176 TWh), and projected 6.7% to 12% by 2028. The International Energy Agency's 2025 Energy and AI report put global data-centre use at about 415 TWh in 2024, projecting about 945 TWh by 2030.

The technology companies have signed up. Microsoft's 20-year deal (September 2024) is restarting Three Mile Island Unit 1, renamed the Crane Clean Energy Center, helped by a $1 billion federal loan in November 2025. Google agreed in October 2024 to buy up to 500 MW from Kairos Power's small advanced reactors by 2035. In June 2025, Amazon contracted for power from Pennsylvania's Susquehanna plant until 2042, and Meta for 20 years from the Clinton plant in Illinois.

Restarts are harder than announcements. Michigan's Palisades plant, closed in 2022, was meant to restart by the end of 2025. Fuel loading began in late August 2026 and was paused when a fuel assembly tilted; as of late October there was no new date.

Governments have joined in. In May 2025, President Trump signed executive orders aiming to quadruple US nuclear capacity to 400 GW by 2050. A Declaration to Triple Nuclear Energy by 2050, launched at the COP28 climate summit in Dubai at the end of 2023, had 39 national signatories by October 2026.

Small modular reactors (SMRs) are the industry's answer to its cost problem. The idea is to build reactors of up to about 300 MW in factories, in series, so that costs fall with repetition. The status is early:

• China's HTR-PM, a small gas-cooled plant, entered commercial operation in December 2023. Its Linglong One (ACP100), under construction since July 2021, was expected to begin commissioning during 2026.
• Canada is furthest along in the West. Ontario Power Generation received a construction licence for a GE Hitachi BWRX-300 at Darlington in April 2025, poured the first nuclear concrete on 26 June 2026, and expects the first unit to run in 2030. Four units are estimated at C$20.9 billion.
• NuScale in the US had a 50 MW design certified by the NRC in January 2023 and a larger 77 MW version approved in May 2025. But its first planned plant, in Idaho, was cancelled in November 2023 after its estimated cost rose to $9.3 billion.
• TerraPower received an NRC construction permit in March 2026 for its Natrium plant at Kemmerer, Wyoming: the first permit in over 40 years for a commercial power reactor that is not a light-water reactor.
• Britain chose Rolls-Royce SMR in June 2025 and signed a contract in April 2026 for three units at Wylfa in Wales, with power expected by the mid-2030s.

The SMR pitch: factories will do for reactors what they did for cars. No one has yet built enough to prove it.

Part 7 — Fusion: the other nuclear

Fission splits heavy atoms. Fusion joins light ones, the reaction that powers the Sun. Its fuel is hydrogen isotopes, it cannot run away the way a fission chain reaction can, and it produces far less long-lived waste. The catch is that it needs temperatures of about 150 million °C and a way to hold the fuel together long enough. ITER points out that fusion is not a chain reaction: if anything goes wrong, it simply stops. Fusion has been "thirty years away" for most of the last seventy. Two real milestones have made the joke a little less true.

Ignition, December 2022. On 5 December 2022, at Lawrence Livermore National Laboratory's National Ignition Facility (NIF) in California, lasers delivered 2.05 megajoules of energy to a tiny capsule of fuel, and the fusion it ignited released 3.15 megajoules. For the first time, a fusion experiment produced more energy than the energy put into the fuel. By June 2026 NIF had achieved ignition eleven times, and its record, set on 7 April 2025, was 8.6 MJ from 2.08 MJ of laser light, a gain of about 4.

The fine print matters. The lasers themselves drew roughly 300 MJ of electricity to produce that 2.05 MJ pulse (according to figures reported at the time), so the facility as a whole used about a hundred times more energy than the fusion produced. Ignition proved the physics, not a power station.

Fusion's waste problem is also smaller. ITER says a fusion plant would produce no high-activity, long-lived waste; its main waste would be machine parts made radioactive by neutrons, most of which could be recycled within about 100 years. That is a real advantage over fission, if fusion can be made to work at all.

ITER, the international machine under construction in southern France, is the main public bet on the other approach, holding plasma in magnetic fields in a doughnut-shaped tokamak. Its members are China, the European Union, India, Japan, Russia, South Korea and the United States. Its aim is a gain of 10. Its schedule has slipped badly. Under the baseline revised in 2024, after defects were found in key components, research operation begins in 2034 and deuterium–tritium fusion in 2039, four years later than planned in 2016. Its original construction estimate was about €5 billion. Later estimates run several times higher, and ITER itself says the cost is being reassessed.

Private fusion is the newer story. The Fusion Industry Association says more than $15 billion has been invested in private fusion companies. Commonwealth Fusion Systems says its SPARC machine in Massachusetts will reach Q above 1 in 2027. Helion has agreed to sell electricity to Microsoft from 2028 and began building a plant at Malaga, Washington, in July 2025. These are company targets, not results. Fusion's history is full of confident dates, and the honest position is that a demonstration of net electricity in the late 2020s or early 2030s would be a major achievement. Fusion supplying a large share of the world's power before 2050 would be a surprise.

Part 8 — The strongest case for, and against

Here are both cases, as their best advocates would put them.

The case for. Nuclear power is the only proven source of very low-carbon electricity that runs day and night regardless of weather, in large amounts, on a small footprint. France, and Sweden alongside its hydropower, decarbonised their grids with it decades ago. Measured by deaths per unit of energy it is among the safest sources ever used, alongside wind and solar and far ahead of fossil fuels, even when Chernobyl and Fukushima are counted. Its waste is small in volume, tracked, and now has a licensed permanent solution in Finland. Its fuel is so concentrated that a country can stockpile years of it. A grid run largely on wind and solar needs firm power to fill windless winter weeks, and nuclear power does that without burning anything. The costs have been high because the West stopped building and forgot how. Where countries build steadily and to standard designs, as France did in the 1980s and China and South Korea have since, costs fall. Abandoning working reactors, as Germany did, meant burning more coal.

The case against. New nuclear plants in the West have been, again and again, years late and billions over budget: Vogtle, Flamanville, Hinkley Point C, and Finland's Olkiluoto 3, which was due in 2009 and began regular production in 2023. Meanwhile solar and wind have become, in the investment bank Lazard's 2026 analysis, the cheapest new sources of electricity without subsidies, and they are quick to build. Every euro and year spent on a reactor is a euro and year not spent on cutting emissions faster. Accidents are rare but, when they happen, they empty regions for decades, and Fukushima showed that even a wealthy, technically capable country can be caught out. Spent fuel needs safeguarding over timescales no institution has ever managed, and most countries still have nowhere to put it. Civilian nuclear technology has been a path to weapons. Reactors in a war zone are now a reality, at Zaporizhzhia. SMRs and fusion are promises, and the industry has a seventy-year history of promises, starting with "too cheap to meter".

What is genuinely unsettled is mostly empirical, and much of it could be answered within a decade:

Both cases agree on more than they admit: that burning coal is the deadliest way to make electricity, that Western nuclear construction has been badly managed, and that the waste is small in volume. They part ways on cost, on time, and on how much a rare disaster should count.

Then there is a value question that evidence won't settle: how to weigh a small chance of a large disaster against the steady, invisible harm of burning fossil fuels. France and Germany answered that question differently, and both answers were reasonable.

The takeaway

Born in the Manhattan Project and sold as "Atoms for Peace" in 1953, nuclear power grew twenty-five-fold between 1970 and 1990, then stalled on cost, protest and accidents. Three Mile Island (1979) killed no one. Chernobyl (1986) killed about 30 people in the first weeks and caused more than 6,000 childhood thyroid cancers; projections of its long-term toll run from about 4,000 to over 90,000 deaths, depending mainly on how low doses are modelled. At Fukushima (2011) the evacuation killed far more people than the radiation. Per unit of energy, nuclear is among the safest sources, with wind and solar; its weakness in the West has been cost and construction time. As of October 2026, 417 reactors operate, 78 are being built (37 in China), Finland has licensed the first permanent waste repository, tech companies are buying nuclear power, and SMRs and fusion remain mostly promises with dates attached.

Try this: look up your own country's electricity mix on Our World in Data (search "electricity mix" and your country). Find the share from nuclear, coal and gas, then multiply each source's annual TWh by its deaths per TWh in the table in Part 3. You'll get a rough estimate of how many lives each part of your electricity costs a year. Then decide for yourself which risks you would trade for which.

Sources & Further Reading

Sources
- Chicago Pile-1: Atomic Heritage Foundation; Wikipedia, "Chicago Pile-1" (sources disagree on the first run's length, so "minutes"). Manhattan Project: US DOE Office of Legacy Management. Hiroshima and Nagasaki estimates: Wikipedia (lead).
- Atoms for Peace: Eisenhower Presidential Library; criticism summarised from Wikipedia, "Atoms for Peace". "Too cheap to meter": Wikipedia article of that name.
- First reactors, Rickover, Yankee Rowe, 1970s cancellations: World Nuclear Association (WNA), "Outline History of Nuclear Energy" (updated 15 March 2026); Wikipedia on Obninsk, Calder Hall and Shippingport. PWR count: WNA, "Nuclear Power Reactors" (5 August 2026). R. Cowan, "Nuclear Power Reactors: A Study in Technological Lock-in", Journal of Economic History (1990).
- Generation, shares and carbon intensity, 1970–2025 (France, Germany, Sweden, world) and per-capita electricity: Our World in Data (Ember and Energy Institute data), downloaded 31 October 2026.
- Three Mile Island: US NRC backgrounder (reviewed 10 September 2026); WNA (23 March 2026); Wikipedia (evacuation, film and clean-up details).
- Chernobyl: WNA (15 April 2026); WHO Chernobyl Q&A (UNSCEAR's 6,000+ thyroid cancers, 15 fatal; IARC's 16,000); WHO news release of 5 September 2005 (Chernobyl Forum). Other estimates: Wikipedia, "Deaths due to the Chernobyl disaster". UNSCEAR's 2012 low-dose statement, as quoted in Wikipedia, "Linear no-threshold model" (UNSCEAR's site could not be reached). Cancer as nearly one death in six: WHO fact sheet (2024 data).
- Fukushima: WNA (23 September 2026), including UNSCEAR 2021. Tsunami toll: official 2021 figures, via Wikipedia.
- Deaths per TWh: Hannah Ritchie, Our World in Data, "What are the safest and cleanest sources of energy?" and its chart data (2,747 deaths over 96,876 TWh, 1965–2021).
- France: WNA (1 April 2026); Wikipedia, "Messmer plan" and "Flamanville Nuclear Power Plant". Germany: WNA (8 April 2026); Jarvis, Deschenes and Jha, NBER Working Paper 26598 (2019), Journal of the European Economic Association (2022).
- Waste: US DOE, "5 Fast Facts about Spent Nuclear Fuel" (2022, including the 90% of energy left and no US recycling); WNA, "Storage and Disposal of Radioactive Waste" (23 October 2026); Posiva press release, 22 October 2026; WNA Finland page (6,500 tonnes, sealing around 2120); Wikipedia on Onkalo and Yucca Mountain.
- 2026 status: IAEA PRIS world and country statistics, read 31 October 2026; WNA world, USA (27 October 2026), UK, China, Canada, Finland and SMR pages (September–October 2026); US EIA, "U.S. nuclear industry" (March 2026); Wikipedia on Vogtle and NuScale; WNA Net Zero Nuclear page on the tripling declaration (2 October 2026); Lazard, LCOE+ 2026 summary page.
- Data centres: US DOE, 20 December 2024 (Lawrence Berkeley National Laboratory report). IEA, Energy and AI (10 April 2025), as reported by S&P Global Commodity Insights the same day (the IEA site could not be reached).
- Fusion: LLNL, "Achieving Fusion Ignition"; the ~300 MJ laser draw as reported by NBC News at the time; ITER FAQ and Wikipedia, "ITER"; Fusion Industry Association; Commonwealth Fusion Systems' SPARC page; Helion's news page.

Go deeper
- 📘 The Making of the Atomic Bomb — Richard Rhodes. The classic history of how fission became a weapon, including Chicago Pile-1.
- 📘 Energy and Civilization: A History — Vaclav Smil. Nuclear power in the long story of how humans have powered themselves.
- 📘 Midnight in Chernobyl — Adam Higginbotham. A careful reconstruction of the accident and the Soviet response.
- 🌐 IAEA PRIS: Power Reactor Information System — the official live count of the world's power reactors.
- 🌐 Our World in Data: What are the safest and cleanest sources of energy? — the deaths-per-TWh comparison, with its method.
- 🌐 World Nuclear Association information library — detailed country profiles, updated often. It is the industry's own body: read its framing as one side's case.
- 🌐 NRC: Backgrounder on the Three Mile Island accident — the regulator's plain account.
- ▶️ Into Eternity (2010) — Michael Madsen's documentary filmed inside Onkalo.

Related pieces in this library: How a Nuclear Reactor Works (Physics 14) · Energy: The One Currency of the Universe (Physics 2) · How Electricity Actually Reaches Your Wall (Physics 4) · The Physics of the Greenhouse Effect (Physics 15) · Marie Curie: Two Nobel Prizes, One Radical Idea (History-Makers 3) · France: A Condensed History of Revolution, Republic, and Grandeur and Germany: A Condensed History of Fragmentation, Catastrophe, and Repair (Condensed History of Nations 8 and 5) · The 1950s (Decades of Leadership 4, on Eisenhower and the Cold War).