In the end, the accident resulted in no injuries or loss of life. Two decades later, The Economist visited the Pennsylvania hinterlands and found the second, unproblematic reactor still running well and enjoying strong local support. It cranked out power until it was mothballed in 2019 owing not to safety concerns but to competition from cheap shale gas.
Now Three Mile Island is coming back from the dead. On September 20th Microsoft, a tech giant, and Constellation Energy, the utility that decommissioned the trouble-free reactor, signed a deal to return it to service. The utility will spend about $1.6bn to restore the plant by 2028. Microsoft will then buy its carbon-free power for the next 20 years.
Having plunged through the 1990s and 2000s, the share of global nuclear power generated by new plants is now rising again (see chart 1). Although America is home to 94 conventional nuclear reactors, about a fifth of the world’s total, it has built few in recent decades.
There are, however, over five dozen nuclear reactors under construction worldwide, mostly in China and Russia but increasingly in other places, too (see chart 2). In July, for instance, the Czech Republic finalised plans for a $17bn nuclear project. And interest in small modular reactors (SMRs), which are cheaper and quicker to build, is surging everywhere. A new age for nuclear power may be dawning.
Demand from tech giants such as Microsoft, Amazon and Google, all of which are racing to find sufficient clean power to feed the data centres underpinning the artificial-intelligence (AI) boom, is turbocharging the nuclear revival. Brad Smith, president of Microsoft, explains that even before AI came along, his firm’s promise to remove more greenhouse gases from the atmosphere than it has contributed during its entire history was ambitious. Now, thanks to AI, he reckons Microsoft will consume five to six times as much electricity by 2030 as it anticipated when it set the goal in 2020.
Nuclear power is well suited to data centres because it provides a reliable source of clean energy around the clock, unlike wind and solar, notes David Victor of the University of California at San Diego. The vast scale of nuclear-power plants also makes them attractive: running an AI data centre can require as much as a gigawatt (GW) of power. That could be provided by millions of solar panels—or just a single nuclear-power plant.
Unfortunately, there are few decommissioned nuclear reactors in America, Europe and Japan that would be suitable for big tech to resurrect. New plants will therefore need to be built. These projects absorb enormous amounts of capital and are sensitive to interest-rate fluctuations and long delays. What is more, the risk of overruns is higher today because the supply chains and expertise needed to build nuclear power have atrophied, at least outside China and Russia. A case in point is the recently opened Vogtle nuclear plant in Georgia, America’s first in decades. It cost $35bn, more than double the initial estimate, and was completed seven years late.
One way for tech giants to mitigate the risk is to bring in outside investors. Last month Microsoft teamed up with BlackRock, an asset manager, and other investors to launch a $30bn infrastructure fund devoted to AI. Mr Smith, who says the fund’s firepower could rise to $100bn once debt is added, expects it to finance nuclear projects, among other things. He also expects three or four such funds led by other tech firms to emerge, and suggests that Microsoft “may participate in all of them”.
At the same time, tech titans have taken a growing interest in SMRs. Currently there are only two in operation, in China and Russia. But Western manufacturers of power equipment such as GE Vernova and Rolls Royce have been busily developing models of them, as have numerous startups. TerraPower, an SMR startup backed by Bill Gates, broke ground on its first plant in Wyoming in August. Oklo, another startup which counts Sam Altman, the boss of OpenAI, among its backers, intends to deploy several small nuclear plants by 2030. Earlier this year it signed an agreement to provide Equinix, a data-centre operator, with 500 megawatts of nuclear power that includes a prepayment that should help it fund construction.
There are more SMRs planned or under construction in America than anywhere else in the world, thanks in large part to enthusiasm from the tech industry. Last month Oracle, a software giant, unveiled plans for a GW-scale data centre to be powered by three SMRs for which the firm has already secured permits. On October 3rd Sundar Pichai, the boss of Google, confirmed that his firm is also looking into SMRs to power its data centres.
Chain reaction
Still, the technology behind SMRs remains unproven at scale. Meanwhile, the hefty and uncertain price tag of building conventional nuclear-power plants will continue to weigh on the industry’s growth—and big tech’s hopes of powering the AI revolution without spewing vast amounts of carbon into the atmosphere. Chris Womack, the chief executive of the Southern Company, which owns the nuclear plant in Georgia, reckons that public support would help. “The government must provide some cost-overrun insurance,” he argues. Streamlining lengthy permitting processes would be even better.
Governments certainly seem to be warming towards nuclear power, after decades of frostiness. At an event held during UN Climate Week in New York last month, officials lined up to reaffirm a commitment made by over 20 countries at the cop climate summit in Dubai last year to triple the global output of nuclear power by 2050. Fourteen big banks also said they are ready to finance new nuclear projects. Pointing to the surge in demand from AI, Mohamed Al Hammadi, head of the Emirates Nuclear Energy Corporation, declared: “We have witnessed a step change in momentum across the nuclear sector.” For it to continue, a mix of technical ingenuity, financial wizardry and supportive policy will be needed.