Nuclear power is experiencing an unexpected renaissance, driven not by energy policy but by artificial intelligence's insatiable appetite for electricity. Tech companies like Microsoft, Google, Amazon, and Meta have discovered that AI data centers require the kind of steady, carbon-free power that only nuclear reactors can provide around the clock. Rather than wait years for new reactor designs to be built, companies are now pursuing a faster strategy: restarting decades-old plants that were mothballed for economic reasons.

Why Are Tech Giants Turning to Nuclear Power for AI?

The answer lies in the fundamental difference between how AI systems consume electricity and how renewable energy sources produce it. AI data centers run continuously, pulling enormous, steady loads every hour of every day. Wind turbines and solar panels are economical, but they generate power only when weather conditions permit. A nuclear reactor, by contrast, operates at full capacity regardless of time of day or season, delivering the reliable baseload power that AI infrastructure demands.

This mismatch between AI's constant hunger and renewables' intermittent supply has created a unique market opportunity. Microsoft's deal with Constellation Energy exemplifies the trend. In September 2024, the software giant committed to purchasing the entire output of Three Mile Island Unit 1 for the next twenty years, securing all 835 megawatts of electricity to power its data centers across the regional grid operated by PJM Interconnection.

What Makes Three Mile Island's Restart Different From Building New Reactors?

Three Mile Island Unit 1 is not the reactor that failed catastrophically in 1979. That was Unit 2, which suffered a partial meltdown and never operated again. Unit 1 sat adjacent to the accident site, operated safely for four decades afterward, and was shut down in 2019 only because cheap natural gas made it unprofitable to run. Now, with a guaranteed buyer in Microsoft, the economics have shifted dramatically.

Restarting an existing reactor is substantially faster than building new nuclear capacity from scratch. Constellation Energy is investing $1.6 billion to bring Unit 1 back online, renamed the Christopher M. Crane Clean Energy Center in honor of former Exelon chief executive Chris Crane. The company has already ordered three new main power transformers, one of the largest single equipment purchases the restart requires, and accelerated their delivery to the site in 2026.

The federal government is backing the effort. In November 2025, the Department of Energy closed a $1 billion loan for the project, allowing Constellation to borrow at lower rates and reinvest the savings into plant upgrades. The restart is expected to bring more than 600 jobs back to Dauphin County, Pennsylvania, and Constellation now projects the reactor will return to service in the second half of 2027, a year ahead of its original 2028 timeline.

How to Understand the Regulatory Path Forward

• NRC Environmental Review: The Nuclear Regulatory Commission released a draft environmental assessment in June 2026, with a preliminary finding of no significant environmental impact. The public comment period runs until July 8, 2026, before the agency makes its final decision on whether to approve the restart.
• Grid Transmission Challenge: PJM Interconnection initially determined that the regional transmission network needed major upgrades, including new 765- and 500-kilovolt power lines, before it could safely carry all 835 megawatts. Those upgrades were not expected until December 2030, creating a timing problem for the reactor.
• FERC Waiver Solution: On June 1, 2026, the Federal Energy Regulatory Commission approved a waiver allowing Constellation to shift 760 megawatts of grid-connection rights from its Eddystone power plant near Philadelphia to the Three Mile Island restart. This permits the reactor to deliver its full output on the 2027 timeline without waiting years for transmission infrastructure.

The regulatory momentum is clearly running in favor of the restart. Energy Secretary Christopher Wright toured the station in December 2025 and framed the project as "delivering on two of our large promises," referring to cheaper electricity and winning the race in artificial intelligence.

Is the U.S. Strategy Spreading Internationally?

The nuclear-AI power strategy is not limited to the United States. Across the Atlantic, a consortium led by Polish billionaire industrialist Michał Sołowow announced plans to build 14 small modular reactors on three sites across the United Kingdom, representing a £35 billion private investment.

Sołowow's company, SGE, plans to deploy enough small modular reactors (SMRs) to power the equivalent of 8 million UK homes for more than 60 years. Each 300-megawatt reactor would cost between £2.2 billion and £2.5 billion to build, using the GE Vernova Hitachi BWRX-300 design, a tenth-generation boiling water reactor developed jointly by the U.S. manufacturer GE Vernova and Japanese industrial conglomerate Hitachi.

The UK consortium has submitted an application to use the Oldbury site in south Gloucestershire, a former nuclear plant location earmarked under the government's advanced nuclear framework. Notably, Google Cloud has signed a joint venture agreement with SGE, with plans to invest up to £4.5 billion in data centers to use the nuclear output, though this is viewed as an accompanying proposal separate from the current application.

"Because of this, I am confident we will set a new standard for nuclear development by combining our disruptive business model with the BWRX-300's 10th-generation proven technology. We will rely strongly on the UK supply chain; it is a critical element for our project," said Michał Sołowow.

Michał Sołowow, Founder and CEO of SGE

The UK Labour government has actively promoted this strategy. Prime Minister Keir Starmer called for tech companies to work alongside the government to build SMRs to power energy-intensive AI data centers across Britain. SGE's plans put it in competition with Rolls-Royce, which won a government competition earlier this year to start generating power by 2032 at the earliest.

Unlike the controversial Sizewell C project, which uses a model where developers are paid during construction and billpayers risk larger costs if delays occur, SGE has opted for a contracts for difference scheme similar to Hinkley Point C. This approach guarantees a fixed rate from energy bills once the project begins generating electricity, shifting financial risk away from consumers.

Why Are Existing Reactors Becoming More Attractive Than New Builds?

The shift toward restarting mothballed reactors reflects a pragmatic calculation. New reactor construction in the United States faces years of regulatory review, design challenges, and construction delays. Small modular reactors, while promising, do not yet exist commercially in the U.S. market. By contrast, Three Mile Island Unit 1 is a proven design that operated safely for decades and requires only equipment upgrades and regulatory approval to restart.

This strategy is spreading. In Michigan, the Palisades plant on the shore of Lake Michigan has already become the first U.S. reactor pulled out of decommissioning and cleared to run again. A shuttered plant in Iowa is attempting to follow the same path. Reviving dead reactors has quietly turned into a strategy for meeting AI's power demands faster than new construction could deliver.

The convergence of AI's energy needs, tech companies' willingness to sign long-term power purchase agreements, and regulatory agencies' openness to reactor restarts has created a new market dynamic. Nuclear power, once viewed as economically uncompetitive against cheap natural gas, is now being valued for its ability to provide the steady, carbon-free electricity that AI infrastructure demands. Whether this trend accelerates or faces obstacles in the coming years will depend on how quickly regulators approve pending applications and whether new reactor designs eventually reach commercial viability.

" }