How the Lights Stay On

Chapter 10: The Atom Comes Back

If you drive south on Pennsylvania Route 441 along the Susquehanna River, between Harrisburg and Lancaster, you will eventually pass four enormous concrete cooling towers rising out of an island in the middle of the river.

The towers are pale gray. They taper inward at the waist and flare out at the top like wineglasses the size of football stadiums. Two of them have been silent since 1979. The other two have been silent since 2019.

This is Three Mile Island.

For most Americans alive today, the name itself means an accident. On March 28, 1979, one of the reactors at this plant — Unit 2 — suffered a partial meltdown. A combination of a stuck valve, a faulty instrument reading, and operator confusion led to part of the reactor core melting inside its containment vessel. The accident did not kill anyone. The amount of radiation that escaped to the surrounding area was, according to most studies done since, not enough to cause measurable health effects. But the country’s faith in nuclear power did not survive intact.

In the days after Three Mile Island, the American nuclear industry stopped. Power plants under construction were canceled. Plants on the drawing board were abandoned. Public opinion turned against new reactors. Anti-nuclear protests grew. For the next four decades, the United States built almost no new nuclear power.

The plant’s other reactor, Unit 1, was unaffected by the 1979 accident. It kept running, safely, for another forty years. In 2019, its owners shut it down because it could not compete economically with cheap natural gas. The cooling towers went cold.

This chapter is about why those towers, in the spring of 2026, are getting ready to come back to life.

On September 20, 2024, the energy company Constellation announced a deal that would have seemed impossible a few years earlier.

Constellation owned Three Mile Island Unit 1 — the reactor that had been shut down five years before. They had signed a contract with Microsoft. A twenty-year contract. Under the contract, Constellation would spend roughly $1.6 billion restoring Unit 1 to working condition. Microsoft would buy every kilowatt-hour of electricity the plant produced, for twenty years.

The reactor would be renamed the Crane Clean Energy Center, after a former Constellation CEO. It would have a planned restart in 2027 or 2028, pending regulatory approval. Its operating license would be extended through at least 2054.

The reason Microsoft wanted nuclear power, specifically, was its data centers. Microsoft is in an arms race with Google, Amazon, and a handful of other companies to build out the infrastructure that runs the world’s artificial intelligence. Each new data center can use as much electricity as a small city. And Microsoft, like its competitors, has made public commitments to power its operations with carbon-free electricity, twenty-four hours a day, every day of the year.

Solar and wind cannot do this on their own. They generate when the weather lets them, not when the data center needs them. Microsoft could buy enough renewable energy on average to cover its annual usage, but during the hours when the sun was not shining and the wind was not blowing, the data center would still be drawing power from the grid — which would still, in most places, be partly fossil-fueled.

Nuclear is different. A nuclear plant runs at full power for nine years out of ten. It produces almost no carbon emissions over its operating life. It is the only proven, scalable source of electricity that is both around-the-clock and low-carbon, on the timescale Microsoft and its competitors need.

So Microsoft did the math, and decided to pay for a closed nuclear plant to come back.

In November 2025, the Trump administration announced a one-billion-dollar federal loan to support the project. The plant is now expected to come back online sometime in 2027 or 2028.

Three Mile Island, the site whose name once stopped American nuclear power in its tracks, is one of the reasons American nuclear power is restarting.

The Crane Clean Energy Center is not the only deal of its kind. It is one of about a dozen significant nuclear announcements in the last three years, and it is part of a much larger shift.

Three pressures have been building up at the same time.

First, the demand for electricity is exploding. For nearly twenty years, US electricity demand was almost flat — efficiency gains canceling out growth. That changed around 2022. Data centers, electric cars, heat pumps, reshored chip and battery factories — all of them want more electricity, all at once. Total US demand is projected to grow 25 percent by 2030. Some of the new demand is so concentrated geographically (Northern Virginia, central Ohio, eastern Texas) that local grids are already straining.

Second, climate goals require firm clean power. Cleaning up the grid using only solar, wind, and batteries is harder than it looks. The intermittency problem from Chapter 6 means you need enormous amounts of either storage or backup, both of which are expensive at scale. A growing number of engineers and policymakers have concluded that getting to a clean grid almost certainly requires firm clean sources too — sources that run regardless of weather. The two main candidates are nuclear and enhanced geothermal. Geothermal is promising but small. Nuclear is proven but politically and economically scarred.

Third, the economics have shifted. Restarting a closed nuclear plant turns out to be much cheaper than building a new one. The plant is already there. The transmission lines are already there. The skilled workforce can be rehired. The license is renewable. For about one to two billion dollars, you can bring back almost a gigawatt of zero-carbon, round-the-clock electricity in three or four years. Compared with the cost and timeline of building a new reactor — we will get to those numbers in a moment — restarts are a bargain.

All 18 chapters of How the Lights Stay On · Library · All courses