Why Is Nuclear Suddenly Back?

On June 18, 2026, in the Utah desert, a small nuclear reactor came to life for the first time.

It had been flown in from California a few months earlier on three C-17 military cargo jets. The machine is called Ward 250, and it's named after a man: the builder's great-grandfather, a physicist who worked on the Manhattan Project. The builder himself is 27 years old. He dropped out of school at 16, taught himself to code, started two software companies, and then decided the family business was unfinished. His startup is now worth about two billion dollars.

His reactor wasn't alone. It was racing a deadline the White House set on purpose: get new American test reactors running by the Fourth of July, the country's 250th birthday. Four made it. One went from breaking ground to a live chain reaction in under eight months. The last crossed the line at twenty minutes past midnight on Independence Day itself. To be precise, these were low-power test runs, first sparks rather than power plants. But America hadn't seen a spark like that in a very long time.

And here's the part that should stop you cold. At the same moment, in Pennsylvania, workers are rebuilding the most infamous name in American energy: Three Mile Island. The plant your generation was taught to fear is being switched back on, under a twenty-year contract, with power expected next year. And the customer paying for it isn't a utility. It's Microsoft.

So that's the question on the door. Nuclear was supposed to be dead, dangerous, and buried. Now the richest companies on earth are fighting to buy it, and Washington is clearing the road. What do they know that you were never told?

Strip away sixty years of movie music and a nuclear plant is a kettle with a very strange flame. In the core sit bundles of small ceramic pellets of uranium. When a slow neutron hits a uranium atom, the atom splits, throws off heat and a few fresh neutrons, and those neutrons split more atoms. A chain reaction. The water flowing past does two jobs at once: it carries the heat away, and it slows the neutrons to the speed where they split atoms best. In most American reactors, if the water is lost, the reaction chokes itself out. The physics fails safe.

Everything after that is 1890s technology. Hot water makes steam, steam spins a turbine, the turbine makes electricity. That's the whole trick. Split atoms, boil water. The only exotic part is the fuel.

Sit with that one. A pellet you could hide in your hand does the work of a ton of coal. Run the numbers up the chain and it gets stranger: a large reactor powers a city of a million people for a year while converting about one kilogram of matter into pure energy. A city, for a year, on a bag of sugar. And if every watt you use for your whole life came from nuclear, the leftover fuel would fit inside one soda can.

It's also the hardest-working machine on the grid. A modern plant runs flat-out 92 to 93 percent of the hours in a year. Solar manages about 23 percent, wind about 35. Nothing else shows up for work like that.

The man who made it reliable was a stubborn immigrant kid from Poland named Hyman Rickover. He built the Navy's nuclear fleet, and his submarine reactor design became the template for nearly every American power plant. His program's record, across some 273 reactors and 175 million miles steamed since 1955: no reactor accident, and no release of radiation that ever harmed the public. The Navy trusts twenty-year-olds to run these machines under the ocean. That record is why.

America fell in love with this machine fast. The first full-scale plant, at Shippingport, Pennsylvania, went up in 32 months for about 72 million dollars. Across the next generation the country switched on more than a hundred reactors. Then, in 1978, it broke ground on its last one. And stopped. Not slowed. Stopped. Thirty-five years passed before anyone poured concrete for a new American reactor.

What happened in 1978 and '79? A perfect storm, and it's almost too on-the-nose to believe. On March 16, 1979, Hollywood released The China Syndrome, a thriller about a reactor meltdown that could render "an area the size of Pennsylvania" uninhabitable. Twelve days later, a reactor at Three Mile Island, in Pennsylvania, partially melted down. The fear had a trailer before it had a fact pattern.

Here's what the accident actually did, measured by every agency and university that studied it for decades: the average neighbor received about 1 millirem of extra radiation. A chest X-ray is about six times that. Confirmed deaths: zero. The worst nuclear accident in American history, and the physical harm to the public was less than a sixth of a chest X-ray.

But nobody measured the fear in millirem. The rules piled on until the numbers stopped working. Even before the accident, a decade of tightening had already added 41 percent more steel, 27 percent more concrete, and 36 percent more cable to the same plant. Reactors caught mid-construction when the accident hit were finished at a median of 2.8 times their original cost. The safety standard became "as low as reasonably achievable," which sounds noble until you see what it does in practice: every time builders find a saving, a new rule eats it, forever. And the regulator billed builders by the hour for reviews that took years.

One more receipt, so you know this wasn't the technology's fault: over those same decades, South Korea's nuclear construction costs went down. Same physics. Same machine. Different rules, different story. The freeze was a choice.

The machine never failed. The story about the machine did.

Start with the only fair test: deaths, per unit of energy produced, across every source. Not vibes. Bodies counted by researchers, including the projected long-term Chernobyl estimates and even the Fukushima evacuation, all charged against nuclear's account.

Read the gold bar again. Charged with its very worst days, nuclear kills at the same rate as wind and solar, and about 99.8 percent less than coal. Meanwhile the routine, boring, unfilmed air pollution from burning fossil fuels kills somewhere between 2.5 and 8 million people every year, depending on whose peer-reviewed study you use. Fossil pollution kills more people before lunch on an average Tuesday than nuclear accidents have killed in seventy years. One dam failure in 1975 that almost nobody can name killed roughly a hundred and seventy thousand people, thousands of times more than every nuclear accident combined. It never got a movie.

One more, for the barbecue: a coal plant sends about a hundred times more radiation into its neighborhood than a nuclear plant making the same electricity. Coal contains trace uranium and thorium, and burning it concentrates the stuff and sends it up the smokestack. The plant everyone fears keeps its radiation locked in casks. The plant nobody fears rains it on the county. And all the used fuel America has ever produced, seventy years of it, would fit on a single football field, stacked less than ten yards deep.

So if the numbers were never the problem, why did the fear win? Partly, honestly, psychology. Radiation is invisible, and the world met it first as a weapon. Scientists who study risk found nuclear sits right where human dread peaks. It's invisible. It's strange. It feels like doomsday. The same people who rank nuclear power as the riskiest thing in modern life will happily take an X-ray, the same radiation, because they chose it and they know it.

But it wasn't only human nature, and this is where the tycoon's question earns its keep. Every reactor that never got built was market share for coal, oil, and gas. And the money trail is not a theory. The chief executive of ARCO, an oil company, personally wrote the 200,000 dollar check that launched Friends of the Earth in 1970, which became the flagship anti-nuclear group in the world. In 2017, "grassroots" groups called Citizens Against Nuclear Bailouts appeared in sixteen states pushing to close nuclear plants; reporters traced the whole campaign to the American Petroleum Institute. The German chancellor who signed his country's nuclear phase-out went to work for the Russian gas pipeline company weeks after leaving office. And when New York closed its Indian Point nuclear plant "for the environment," it was replaced, almost megawatt for megawatt, by three new gas plants. Emissions went up.

Notice what I'm not saying. I'm not handing you a smoke-filled room where villains planned it all. It's quieter than that, and more useful to understand: the fear was cheap to spread, its costs were invisible and slow, and the industries that won never had to lie loudly. They just had to keep the scariest story on stage. Fifty years of energy policy got written by a horror movie's marketing department, and the bill was paid in carbon, in bodies, and in your electric bill.

Fear kept the machine buried for two generations. So what changed? Simple: demand came back, and it came back hungry. American electricity use grew about 0.1 percent a year from 2005 to 2019. Flat, for nearly two decades. The whole system got comfortable. Then the AI buildout arrived, the one we walked through in the data centers file, and the government's own forecasters now expect the strongest four years of demand growth since 2000. The grid needs an enormous amount of new power that runs all day, every day, in the same spot, for decades. Go back to the machine chapter and ask: what runs 92 percent of the year and never blinks?

The buyers noticed before the public did. In the span of about eighteen months, every single one of the tech giants bought nuclear power. Microsoft signed the twenty-year Three Mile Island deal. Google is paying to restart a shuttered plant in Iowa. Amazon parked its data centers next to a Pennsylvania reactor and locked up its output through 2042. And this past January 9th, Meta signed with three nuclear companies for up to 6.6 gigawatts in a single announcement:

Here's a detail I enjoy, one board reading another. The first movers on stranded nuclear power weren't the tech giants at all. They were Bitcoin miners, who quietly set up next to that same Pennsylvania plant years ago, hunting the cheapest reliable power on earth. Then the AI companies showed up with deeper pockets and outbid them for the same electrons. Amazon bought the miners' campus for 650 million dollars. The smart money keeps finding the same answer from different directions: cheap, dense, always-on energy is the scarcest asset of the digital age.

Washington flipped just as hard. The White House ordered the country to build four times its nuclear power by 2050, and cut reactor licensing from roughly five years to eighteen months, and fast-tracked the test program that lit four brand-new reactors for the first time by this Fourth of July. Wyoming broke ground this spring on the first truly new kind of American commercial reactor in a generation. The restarts are moving too: Palisades in Michigan is in final testing for first power this year, and eleven days ago a federal judge threw out the last lawsuit standing in its way. Three Mile Island follows in 2027, Duane Arnold in Iowa in 2029. After nearly half a century, the question in the industry is no longer whether to build. It's how fast.

The fear was priced into everything. The correction of the fear is the opportunity.

Zoom all the way out, because this is bigger than an American story. China is building roughly half of all the nuclear reactors under construction on the planet, about 36 of them at once, and has approved ten or more new units a year for four straight years. On the current pace it passes America as the world's biggest nuclear power around 2030. France, which quietly ran its whole grid about 70 percent nuclear for decades while its neighbors argued, just committed to six more large reactors. Even the World Bank, which hadn't funded a nuclear project since 1959, lifted its ban last summer. Thirty-one countries have pledged to triple global nuclear capacity by 2050. The freeze was a Western choice. The thaw is global.

Now the part the headlines skip: the fuel. Uranium's price has roughly tripled since 2020, from around 32 dollars a pound to the mid-80s, briefly breaking 100. About four of every ten pounds mined on earth come out of Kazakhstan. And America, which banned Russian enriched uranium after the Ukraine war, has until January 2028 before the last waivers expire, with Western fuel plants racing to fill the gap. Every buildout has a chokepoint. The last file taught you that the AI buildout's chokepoint moved from chips to power. The power buildout has a chokepoint too, and it's the fuel chain. Same lesson, one level deeper.

And one honesty check, because we don't do hype here. The small modular reactors you hear about, the venture-backed startups, the fancy new designs: none of them delivers commercial power today. The first arrive around 2028 to 2030 if things go well. The megawatts that actually feed the grid this decade come from restarting and upgrading the big plants we already have, which is exactly where the smart money went first. The frontier is real. It's also not here yet, and anyone selling you certainty about it is selling.