Opinion: The Irony of Powering AI on Atomic Energy

Should the U.S. revive nuclear power to satisfy the growing electricity demands of artificial intelligence?

3 Comments

The former visitor center for the Three Mile Island nuclear power plant, or TMI, was overrun by greenery when I conducted an interview there in 2020 for my graduate research. The ivy had climbed to the second-story observation deck, once a carefully crafted vantage point for visitors to look across the street at the power plant and onto the atomic future it heralded. 

On March 28, 1979, TMI’s Unit 2 reactor suffered a catastrophic loss of coolant, resulting in a partial meltdown that shook public confidence in the growing nuclear power industry. In 2019, the remaining Unit 1 reactor began its 60-year decommissioning. But on Sept. 20, 2024, Microsoft announced its deal with Constellation Energy, the current owner of the plant, to restart Unit 1 by 2028 to power the technology giant’s data centers, many of which are devoted to artificial intelligence.

On the surface, nuclear power offers a seemingly reliable, carbon-free energy source to support AI’s voracious energy demands. But belying that proposed solution is the irony of reviving a dying and dangerous 20th century technology to power a 21st century technology that also promises miracles and has high environmental and human costs.

I had been to the visitor center many times from 2019 to 2023 when I was an anthropology Ph.D. student researching the impact of TMI’s decommissioning on Middletown, Pennsylvania, the small community that hosts the plant. In September 2019, I huddled with a handful of journalists, antinuclear activists, and residents at the funereal press conference that marked TMI’s closing. We listened to the utility’s senior vice president and chief nuclear officer lament the socioeconomic and environmental loss from the shutdown.

The speech mourned the lost promise of nuclear energy, indirectly drawing on the atom’s 1953 rebranding at the height of the Cold War. In his “Atoms for Peace” speech to the United Nations General Assembly that year, President Dwight D. Eisenhower rhetorically transformed the atom from a global weapon of war into an abundant source of energy. But Eisenhower and other experts believed the atom’s potential extended beyond the energy sector. In 1957 at the ratification of the International Atomic Energy Agency, Eisenhower declared that “the splitting of the atom may lead to the unifying of the entire divided world.”

On the surface, nuclear power offers a seemingly reliable, carbon-free energy source to support AI’s voracious energy demands.

Its mid-century proponents described atomic power with almost breathless excitement. In a 1970 article for the Bulletin of the Atomic Scientists, chemist Glenn Seaborg, then chairman of the U.S. Atomic Energy Commission, started by describing a hypothetical speech taking place in 1995. In this imagined future, he wrote that “the atom’s energy used in infinite variety is the lifeblood of our society,” enabling technological miracles in medicine and space travel.

Seaborg had foreseen an atomic utopia just within Americans’ grasp. But seared into many minds in the post-war era were images of nuclear apocalypse. To “reeducate” the American public on atomic energy in the service of peace and prosperity, the U.S. coordinated a mid-century propaganda program.

As the starting point for tours, TMI’s visitor center played a key role in this rebranding. Tours were not only educational. They were a public relations tool, designed to convince a concerned public that nuclear technology was safe — especially after the plant’s 1979 accident.

In 2023, OpenAI’s CEO Sam Altman underwent a public relations tour of another kind. Traveling across Europe, the Middle East, Asia, and Australia, Altman discussed the political and economic consequences of AI with heads of state as well as with AI users. Altman has pointed out the geopolitical consequences of America lagging behind on AI, much like the preoccupations in both the U.S. and the Soviet Union that animated the nuclear arms race of the mid-20th century. Altman believes that AI can be used “to benefit humanity.” Bill Gates has spoken of the technology’s ability to democratize education and medicine across the world. Microsoft was one of the first major companies to invest in OpenAI, pouring in $13 billion over the past five years.

But neither Microsoft nor OpenAI has provided satisfying answers to increasingly unsettling questions about the technology’s less virtuous capabilities, from generating deepfakes to bioterrorist attacks. At the same time, chilling statistics lay out how AI may impact the global economy and the planet. The Pew Research Center found that 19 percent of workers were in jobs most vulnerable to AI, and a Wells Fargo analysis projected a 20 percent increase in U.S. electricity demand by 2030, in part fueled by the AI boom.

Given the reality of AI’s energy demands on an ever-warming planet, proponents of nuclear power see it as a carbon-free solution. But the mining, milling, and production of nuclear fuel, as well as the construction and decommissioning of nuclear plants, emit greenhouse gases at levels ranging from 10 to 130 grams of carbon dioxide per kilowatt hour of power — lower than fossil fuels but higher than wind and hydroelectricity (and roughly on par with solar).

Papering over these economic and environmental realities, Altman and Gates, like Eisenhower and Seaborg, foresee a future in which American technology will create a better, more unified world — the same scientific optimism that powered the peaceful atom. But particularly at TMI, there is another way to remember “Atoms for Peace”: as a cautionary tale of technology outpacing expertise.

Neither Microsoft nor OpenAI has provided satisfying answers to increasingly unsettling questions about the technology’s less virtuous capabilities.

The conventional understanding of TMI’s partial meltdown is that it was contained, a catastrophe narrowly avoided. Government reports indicated a trail of mechanical and human errors. The plant’s pilot relief valve had been open for hours before the control room could correctly diagnose the problem and intervene. In that time, precious cooling water had boiled and evaporated, leaving the reactor dangerously exposed.

The 12-year cleanup of TMI cost $973 million. Grainy black-and-white images from cameras sent into the damaged reactor revealed shocking information: Up to two-thirds of the reactor had been exposed, 45 percent of the uranium fuel had melted, and 19 tons of it had dropped to the reactor floor — all evidence that the accident was closer to nuclear catastrophe than the experts had initially realized.

The people living in the small town that once welcomed TMI as a source of revenue and jobs complained of a metallic taste, burns on their skins, stillbirths among their farm animals, mutated plants, and, decades later, clusters of cancers.

The accident exposed the faults in Eisenhower’s careful distinction between the power plant and the bomb, and the fledgling industry never recovered. When 1995 did arrive, it did not offer the atomic utopia that Seaborg prophesized. Rather than the “lifeblood of society,” the atom seemed almost irrelevant, replaced by lower cost fossil fuels.

Support Undark Magazine

Undark is a non-profit, editorially independent magazine covering the complicated and often fractious intersection of science and society. If you would like to help support our journalism, please consider making a donation. All proceeds go directly to Undark’s editorial fund.

The U.S. is only in the beginning stages of its half-century project to decommission roughly 20 American nuclear plants and dispose of over 90,000 metric tons of nuclear waste from commercial nuclear power reactors, the highest amount of any country in the world. These efforts are underway while regulations loosen over decommissioning plants that continue to house radioactive waste on-site.

Now TMI, in the process of decommissioning, is being revived. What sort of vision of the future would the power plant’s visitor center present to the public in 2028?

Three Mile Island has become a symbol of the U.S. nuclear industry’s overconfidence in its early expertise and of its ultimate failure to deliver on its promises. The irony of it being used to power AI — a technology already notorious for surprising and outpacing its experts — is unavoidable. But there is something even more dangerously ironic about powering the future with the stubborn optimism of the past. Rather than how to meet AI’s voracious energy demands, Three Mile Island might provide a historical vantage to consider whether to power this future at all.


Stephanie Palazzo, Ph.D., is a sociocultural anthropologist and teacher. She is currently writing a book on Three Mile Island’s 1979 accident and present decommissioning.

Republish