Why Did Microsoft Choose a Shuttered Nuclear Plant Over Renewables?

According to Bloomberg Technology, Microsoft's deal with Constellation Energy to restart Three Mile Island Unit 1 is driven by the need for "24/7 carbon-free energy" that wind and solar cannot reliably provide. The U.S. Energy Information Administration (EIA) reported in its 2025 analysis that data center electricity demand could double by 2030, with AI workloads being the primary driver. Microsoft's own sustainability goals require matching every megawatt-hour of consumption with carbon-free generation, a standard that intermittent renewables struggle to meet at scale. The plant, which shut in 2019 for economic reasons, can generate 835 megawatts of baseload power, enough to run roughly 800,000 homes or a cluster of hyperscale data centers. Constellation Energy CEO Joe Dominguez told Bloomberg, "This is the only way to get the volume of clean power Microsoft needs without building new transmission lines for a decade." The choice reveals a stark reality: no amount of solar panels can replace the density and reliability of a nuclear reactor for AI's power profile.

What Are the Real Risks of Restarting a Reactor With TMI's Legacy?

The restart of any nuclear plant faces technical and regulatory peril, but Three Mile Island carries unique baggage. Unit 1 was not damaged in the 1979 partial meltdown of Unit 2, yet public memory conflates both units. Bloomberg reported that the Nuclear Regulatory Commission (NRC) must approve a license transfer and a new operating plan, a process that could take 18 to 24 months. Constellation Energy estimates the restart will cost $1.6 billion, but independent analysts, such as those at the Institute for Energy Economics and Financial Analysis (IEEFA), have noted that nuclear refurbishments routinely exceed initial budgets by 30-50%. A concrete risk: the plant's steam generators and turbine rotors have sat idle for seven years, and replacing major components could push costs beyond $2.5 billion. If the NRC demands additional safety upgrades—such as flood protection or seismic reinforcements—the timeline could slip past 2029, leaving Microsoft's data centers scrambling for backup power from natural gas plants, which would blow its carbon targets.

Who Wins and Who Loses in This Deal?

Stakeholder	Gains	Loses	Verdict
Constellation Energy	20-year captive customer, premium pricing, regulatory momentum for other restarts	Financial risk of cost overruns, reputational risk if delays occur	Clear winner: transforms stranded asset into annuity
Microsoft	Guaranteed carbon-free baseload power, PR win for AI sustainability	Potential power gap if restart is delayed, higher cost than renewables	Short-term winner, long-term gamble
Pennsylvania Ratepayers	None directly; power is for Microsoft, not the grid	May subsidize grid upgrades needed to connect plant to data centers	Loser: bears infrastructure costs without benefit
Renewable Energy Developers	None	Lost opportunity to prove solar+storage could meet AI demand	Loser: nuclear deal crowds out renewable investment
Climate Goals	Large carbon-free generation source	Delay in deploying faster, cheaper alternatives	Toss-up: carbon benefit vs. opportunity cost

Can This Model Scale to Other Retired Nuclear Plants?

Bloomberg noted that Constellation Energy is already evaluating other shuttered plants, including the Duane Arnold Energy Center in Iowa. However, the EIA data shows that only four U.S. nuclear plants have been fully decommissioned and could theoretically be restarted, each facing unique regulatory and physical challenges. The NRC's licensing process for a restart is untested for plants that have been fully defueled, as TMI Unit 1 was. According to nuclear policy expert Dr. Allison Macfarlane, former NRC chair, "Each restart will be a bespoke, multi-year regulatory marathon." The model only works if a hyperscaler like Microsoft or Amazon is willing to sign a 20-year PPA at rates far above wholesale electricity—reportedly $50-60 per megawatt-hour for TMI, versus $20-30 for wind. That premium is a tax on AI that no other industry can afford, making this a one-off solution rather than a template.

1. By Q4 2028, the NRC will demand at least $400 million in additional safety upgrades for TMI Unit 1, pushing the restart cost above $2 billion and the timeline past 2030.
2. Constellation Energy will announce plans to restart a second shuttered nuclear plant (likely Duane Arnold) by mid-2027, but only if Microsoft signs a second PPA at higher rates.
3. By 2029, the U.S. Department of Energy will publish a report concluding that nuclear restarts are 40% more expensive per MWh than a combined solar+storage buildout for AI data centers, creating policy friction.

• AI's energy debt is now a nuclear liability — Microsoft is betting that a 1970s-era reactor can solve a 2020s problem, a mismatch that will create cost and timeline shocks.
• Constellation Energy wins the AI arms race — The company has effectively monetized the AI industry's desperation for carbon-free baseload power, setting a precedent for other nuclear operators.
• Renewables get a cold shoulder — This deal signals that hyperscalers view solar+storage as insufficient, potentially redirecting billions in clean energy investment away from proven technologies.
• Regulatory risk is the unspoken variable — The NRC has not approved a restart of a fully defueled plant; every month of delay increases Microsoft's carbon footprint and operational risk.
• The AI-nuclear marriage is fragile — If TMI's costs spiral, it could poison the well for future tech-nuclear deals, leaving AI companies with no carbon-free path at scale.