The first US atomic rush was a bust. Will Trump’s big nuclear-for-AI plans fare any better?

By Chloe Shrager | July 18, 2025

Amazon's recently acquired data center (foreground) in Salem Township, Pennsylvania, is a stone's throw from the Susquehanna nuclear power plant. In November 2024, the Federal Energy Regulatory Commission blocked the company's request to obtain more electricity directly from the plant. President Trump's nuclear-for-AI executive orders could bypass the traditional market and regulatory regime entirely by deploying reactors on federal sites. (Credit: Talen Energy)

As Big Tech turns to nuclear power to solve the artificial intelligence power problem, critics have cast doubt on energy developers’ ability to build new reactors on a timeline that will satisfy data centers’ energy needs.

High costs and lack of commercial economic viability have been persistent obstacles to new nuclear infrastructure development. But on May 23, President Donald Trump signed four executive orders that represent the most explicit government commitment to nuclear power for artificial intelligence yet.

Three of the orders explicitly mention AI as a driver for nuclear energy development and a potential beneficiary. One directive incentivizes the operation of privately funded advanced nuclear reactor technologies on federal sites—mainly national laboratories or military installations—allegedly to power AI infrastructure, labelled as “critical defense facilities,” and mandates the deployment of small modular nuclear reactors on one of these sites within 30 months.

Previously, tech companies were the most vocal advocates pushing for nuclear power to meet AI’s energy demands. Now the US government—heavily influenced by Big Tech’s hand—has made nuclear power for AI a national security priority, setting a goal of quadrupling the United States’ nuclear capacity from 100 gigawatts to 400 gigawatts by 2050. Whether government intervention can overcome the challenges that have plagued nuclear deployment for decades remains to be seen—and if so, at what cost?

Déjà vu. As with the rise of the nuclear power industry in the 1950s and 60s, the demand for nuclear energy is being created, justified, and incentivized by the government and its national security interests rather than by market forces.

Robert Duffy, a professor of political science at Colorado State University, summarized the history of the US nuclear power industry in a 2004 paper.

“The atomic energy subgovernment was endowed with additional prestige and power because of the program’s identification with national security issues,” Duffy wrote. “The actors in this tightly knit monopoly were united by the conviction that the development of atomic energy, first as a weapon but later as a means of generating electricity, was both necessary and desirable for the nation’s welfare.”

Duffy showed that the government’s rush to create a nuclear industry in the United States ultimately undermined that very industry. The hasty development, government incentives, and ambitious timelines led to cost overruns, safety problems, and public opposition that ultimately killed new nuclear construction for decades.

Today, the Trump administration is repeating history by declaring AI technologies driven by advanced nuclear power generators a key national security interest.

“There seems to be an aspect to the government’s interest in AI which is sort of positing that as the next nuclear weapons race,” Tim Judson, executive director of the Nuclear Information and Resource Service, observes. “If you expect the most powerful countries in the history of the world, and the wealthiest corporations that have ever existed, which are trying to develop […] ‘digital gods,’ to not do everything they can to win that race, then you don’t understand human nature, and you don’t understand geopolitics,” Judson adds, paraphrasing what he heard from a panelist at the Nuclear Energy Institute’s nuclear finance conference in February.

But by trying to rush nuclear power development again for geopolitical reasons (then the Cold War, now the global AI race), the US government risks creating another failed—or at least costly and insufficiently safe—nuclear program.

AI’s unquestioned energy appetite. The government’s nuclear-first approach reflects the staggering scale of AI’s energy demands. Tim Fist, Director of Emerging Technology Policy at the Institute for Progress, says, “you’re roughly looking at tripling the amount of power used for data centers over the next five years, just coming from AI.”

The current global data center annual power consumption sits at around 40 gigawatts of installed electric capacity, but by 2030, he said AI data centers alone will require an additional 120 gigawatts, totaling about 160 gigawatts—just over half of Germany’s entire installed power generation capacity in 2023. Global consulting and auditing firm Deloitte similarly estimates data center electricity demand could rise fivefold in the next decade, reaching 176 gigawatts by 2035.

These massive energy requirements have prompted both the private tech industry and now the federal government to embrace nuclear power as their preferred solution (even as these massive energy requirements are being questioned after China’s DeepSeek showed it was possible to develop and train an AI chatbot at a fraction of the cost of and using 50 to 75 percent less energythan US ones). The Biden administration’s initial executive orders on AI infrastructure development laid the groundwork, but the Trump administration has gone a step further and explicitly targeted nuclear power as the answer to AI’s energy demands.

However, Deloitte’s analysis also predicts that new nuclear power capacity could meet only about 10 percent of the projected increase in data center power demand over the next 10 years. In the United States, where around 70 percent of the world’s most compute-intensive AI models have been trained, this translates to an immediate challenge: “Looking over the next five years, there’s these pretty serious energy requirements,” Fist explains. “90 gigawatts need to be built in the United States over the next 5 years.” That’s about 90 average-size new nuclear reactors. To put that demand in context, as of July 2024, 59 reactors were under construction worldwide, 23 of which were behind schedule.

Before the recent executive orders, the Trump administration had already identified 16 federal sites across the country for data center and energy infrastructure development. The Energy Department’s initiative, following Trump’s other executive orders (“Removing Barriers to American Leadership in Artificial Intelligence“ and “Advancing Artificial Intelligence Education for American Youth”), represents an unprecedented federal commitment to AI adoption and infrastructure.

Tech companies have increasingly aligned with Trump’s nuclear-for-AI strategy.

The appeal is obvious: Nuclear plants provide reliable, nearly 24/7 baseload power that data centers require. Judson notes that tech companies claim they need “the five nines of reliability,” meaning electricity available 99.999 percent of the time all year long. (The average capacity factor for nuclear reactors in the United States was 93 percent in 2023 and 70.7 percent for the 1971–2022 period preceding.)

So far, Amazon is the only major tech company that has made concrete nuclear investments, leasing land at Pennsylvania’s Susquehanna nuclear power plant for a data center that could expand to 960 megawatts of capacity, Judson said. Microsoft and other companies have signed memoranda of understanding for future nuclear power purchases, but these deals—which now align perfectly with the federal government’s nuclear-first strategy for AI—remain theoretical.

The government’s nuclear-AI revolution. The May 23 orders represent the evolution in federal energy policy that began with general energy declarations and ended with explicit nuclear-AI mandates. On his first day in office, Trump signed executive orders declaring a national energy emergency and directing agencies to “unleash America’s affordable and reliable energy and natural resources,” which primarily focused on fossil fuel development but hinted at broader energy expansion.

Subsequent orders explicitly link nuclear expansion to AI competitiveness and national security. As White House Office of Science and Technology Director and former executive in the AI industry, Michael Kratsios, stated in a press release: “Over the last 30 years, we stopped building nuclear reactors in [the United States]—that ends now. Today’s executive orders are the most significant nuclear regulatory reform actions taken in decades.”

Each of the four orders addresses an area designed to accelerate nuclear deployment specifically for AI applications.

The first order directly mandates the development of nuclear power for AI as a national security priority, directing the Army Secretary to establish a program to build a nuclear reactor at a military installation within three years and requiring the Energy Secretary to work with private sector partners to deploy advanced nuclear technology for AI infrastructure within 30 months. These “advanced” technologies notably include small modular reactors (SMRs) and microreactors.

Another order requires a comprehensive overhaul of the Nuclear Regulatory Commission (NRC) to reduce regulatory barriers, decrease dependence on foreign technologies, and require timely reactor licensing decisions—changes designed to speed up nuclear deployment for the energy-hungry AI sector.

Cementing prior Energy Department actions, a third order authorizes rapid nuclear deployment on federal land, with the government offering to lease federal sites for building new energy resources and data centers, streamlining permitting processes without involving private landholders.

Finally, the fourth order seeks to reinvigorate the US nuclear industrial base, including uranium mining expansion, domestic fuel enrichment capacity, and accelerated reactor testing at the Energy Department’s national laboratories.

As a White House fact sheet explains: “The federal government’s advanced computing AI infrastructure will require a substantial increase in scalable power solutions, which advanced nuclear reactors are well-positioned to provide. This will ensure our technological supremacy in the emerging technologies of both AI and nuclear power.”

The most crucial aspect of these orders is how they intend to solve nuclear power’s long-running economic problem: bypassing the traditional market and regulatory regime entirely by deploying reactors on federal sites. This approach would sidestep the Nuclear Regulatory Commission’s traditional oversight and eliminate the need to make a conventional economic case for nuclear power.

According to Duffy, the federal government spent over $1.2 billion to develop reactor technology by 1962—twice the amount spent by private business. One Energy Department study from 1980 concluded that, without federal subsidies, nuclear electricity would have been 50 percent more expensive.

Today’s approach is even more direct. By allowing SMRs to be deployed on federal sites, the government creates an economic case by fiat. The federal government will fund the first fleet of reactors, eliminating the need for private companies to prove commercial viability. As Judson notes about the current situation: “[Nuclear reactor companies] don’t need to ask for subsidies. The government wants to make this stuff happen.”

This strategy echoes the 1950s playbook, when, Duffy noted, “the words and actions of government officials convinced the nuclear industry that in the future there would be nuclear power in America, with or without private sector involvement.”

Sticky problems. Even without the economic hurdles, the fundamental problem remains timing, and presidential orders cannot change the laws of physics. As Mycle Schneider, an independent nuclear policy analyst and main contributor to the World Nuclear Industry Status Report, bluntly states: “I doubt that any SMR would be operating 10 years from now in the Western world.”

Schneider’s skepticism isn’t unfounded. Construction times for nuclear plants average around 10 years, he said, and that’s just the construction phase—which only begins with the pouring of reactor foundations. Even with the Trump administration’s regulatory streamlining and federal site access, the reality of nuclear development timelines clashes directly with AI’s immediate energy needs. “All of these deals with nuclear companies are about future power plants maybe coming online in the 2030s, but all the AI data centers are being built today,” Judson observes.

Small modular reactors have long been promoted by the industry—and now also the government—as a solution to nuclear power’s problems, promising faster construction, lower costs, and standardized designs. The Trump administration’s nuclear orders specifically enable SMR testing and deployment on federal sites, betting that government support can make SMR promises a reality.

But the reality has proven far more complex, even with unprecedented government backing. Canada’s recent approval of the world’s first SMR in a G7 country demonstrates both the promise and the problems. The project’s price tag sits at nearly $21 billion Canadian dollars ($15.1 billion US dollars) for four reactors at Ontario’s Darlington site, roughly $12.5 million US dollars per megawatt—far exceeding the costs of renewable alternatives that can be deployed in a fraction of the time. Even more so, Judson says the energy company GE Vernova-Hitachi chose to pursue its SMR project in Canada because the Canadian Nuclear Safety Commission regulations allow construction permit applications to be submitted with much of the design still incomplete. “The jury is very much still out on whether the BWRX-300 [SMR design] will prove feasible to build on time and on budget, but what we know so far is not encouraging,” he adds.

The long-term management of nuclear waste also poses a sticky issue to new nuclear development, especially the relatively higher waste per gigawatt from SMRs compared to full-scale reactors, which has no permanent solution yet. “If they want to build nuclear power plants at data centers, or build data centers at nuclear power plants, these are sites that are also going to be storing nuclear waste,” Judson points out.

The Trump administration’s aggressive 30-month timeline for nuclear deployment doesn’t give much time to work these issues out. The administration proposes reforms and new U-turn policies for managing and recycling spent fuel storage, but those policies would likely come as new nuclear power plants are already underway.

Without a permanent nuclear waste repository and management policy, every new nuclear facility becomes a de facto long-term storage site. This is particularly problematic for proposed “micro reactors” that would be shipped to sites with fuel already loaded for a pre-determined short period. “Where are these micro reactors going to go after they’ve powered a data center for 10 years?” Judson asks. “Are they going to start piling up dead micro reactors at the data center sites?”

Energy industry impacts of the nuclear-AI boom. Even with the most aggressive government nuclear policy intervention in decades, the disconnect between nuclear industry promises and deliverable reality may continue to widen. Companies like Oklo exemplify this gap, Judson said. Originally marketing a 1.5-megawatt micro reactor, the company’s application was rejected by the NRC as incomplete, he said. Oklo subsequently signed a series of agreements with Switch (an AI technology provider) to build 12 gigawatts of nuclear reactor capacity for the company’s data centers. But as Judson reveals, these “agreements” were power-purchase contracts that Oklo plans to fulfill initially with natural gas generators and potentially transition to nuclear power only if their reactors are ever built and approved.

This pattern reveals a fundamental paradox: Even companies well-positioned to benefit from the Trump administration’s nuclear-friendly policies are hedging their bets with conventional generation sources that can actually be deployed on AI’s timeline.

Fist outlines what he sees as a realistic deployment sequence that challenges the government’s nuclear-first timeline: “I think in the near term, if you want to build fast, it’s kind of solar and battery storage and natural gas. And then, a couple of years after that, you can start bringing on large-scale geothermal online. And then after that, it’s more like small modular reactors.”

The May 23 orders promote this dual-energy solution. One order states that, “In conjunction with domestic fossil fuel production, nuclear energy can liberate the United States from dependence on geopolitical rivals.”

While the Trump administration has made nuclear power central to its AI strategy, it shows less concern for climate considerations than its predecessor. Judson suggests that even in light of new executive orders, this may reduce pressure on tech companies to choose nuclear over fossil fuels: “What I’m seeing is at least some recognition that the power, at least for now, is going to be coming from fossil fuel generation […] and less pressure on the industry politically from the White House to make it seem otherwise.”

The federal government has moved beyond simply supporting AI infrastructure to making nuclear power central to the United States’ AI strategy. But as Duffy’s historical analysis makes clear, government enthusiasm and subsidies don’t guarantee success and can even kill a nascent industry at birth.

The question is no longer whether the government supports nuclear power for AI—it’s whether government intervention can make new nuclear reactors built on AI’s timeline without compromising safety or repeating past mistakes. The answer may determine not just the future of US energy policy, but the United States’ position in the global AI race itself.