Tech Companies Are Becoming Power Companies — And They're Betting Billions on Nuclear
In September 2024, Microsoft and Constellation Energy announced something that would have been unthinkable a decade ago. They were going to restart Three Mile Island.
Not the reactor that melted down in 1979. Unit 1, which had been shuttered since 2019 for economic reasons, had never had a serious accident. Constellation is investing roughly $1.6 billion to refurbish the pressurized water reactor. Microsoft signed a 20-year power purchase agreement. It will buy 100% of the plant's 835-megawatt output — enough electricity to power roughly 800,000 homes. All of it is going to Microsoft's AI data centers.
The facility, now renamed the Christopher M. Crane Clean Energy Center, is roughly 80% staffed as of January 2026, with refurbishments nearing completion. Constellation remains on track for a 2027 restart. But there is a catch. Grid operator PJM has told Constellation the plant may not connect until 2031. That is four years later than planned. The reactor will be ready. The grid won't be.
This is the first time a retired American nuclear plant has been revived for a single corporate customer. The reactor is being restarted because AI needs power. Not the intermittent kind from solar and wind. The constant, reliable, 24/7 baseload that only nuclear can deliver.
The Numbers That Explain Everything
Every major AI company has now signed at least one nuclear deal in 2026. Microsoft, Google, Amazon, Meta — all of them. The scale is enormous.
Microsoft secured 835 megawatts from Three Mile Island through 2047. It's also partnering with Nvidia on an "AI for nuclear" program to accelerate reactor design and deployment.
Amazon has been even more aggressive. The company is working with X-energy on small modular reactors (SMRs). The goal: more than 5 gigawatts online in the U.S. by 2039. It also secured about 2 gigawatts of dedicated nuclear power from Talen Energy. The source: the Susquehanna plant in Pennsylvania. The deal runs 17 years. Amazon acquired the Cumulus data center campus. It is physically connected to the nuclear facility. Amazon can draw up to 960 megawatts without the public transmission grid. The company also tripled its commitment with Energy Northwest. It may deploy 12 SMRs at the Cascade Advanced Energy Facility. Total target capacity: 960 megawatts.
Google signed an agreement with Kairos Power to bring 500 megawatts of advanced nuclear online by 2035. The first small modular reactor is targeted for 2030. In April 2026, Kairos broke ground on the Hermes 2 Demonstration Plant in Oak Ridge, Tennessee. It is the first deployment under the Google deal. Hermes 2 will provide up to 50 megawatts of clean electricity to the Tennessee Valley Authority grid. Google also announced a $1.5 billion expansion of its Alabama campus. The site will eventually be powered by nuclear. That comes through its partnership with Kairos and the Tennessee Valley Authority.
Meta signed agreements with Vistra, TerraPower, and Oklo in January 2026. The goal: 6.6 gigawatts of nuclear power online by 2035 for its Prometheus AI data center in Ohio. That's enough electricity to power the equivalent of about 5 million homes. Under its deal with Vistra, Meta will draw 2,176 megawatts from three nuclear plants. Its agreement with TerraPower will fund two Natrium reactor units. They can generate up to 690 megawatts by 2032. And its deal with Oklo will develop a 1.2-gigawatt nuclear campus in Ohio. Oklo counts OpenAI CEO Sam Altman as one of its largest investors.
A "Bring Your Own Generation" Strategy
What's happening here is a fundamental decoupling of Big Tech from the public utility grid. Companies face aging infrastructure and five-to-seven-year wait times for grid connections. So they adopted what analysts call a "Bring Your Own Generation" (BYOG) strategy. Tech giants are bypassing traditional bottlenecks. They co-locate data centers at nuclear sites. Or they finance the restart of decommissioned reactors.
Tech companies used to buy electricity from utilities. Now they're becoming utilities. Microsoft is restoring a nuclear plant that was mothballed for economic reasons. Amazon is building data centers adjacent to nuclear facilities. Google is funding next-generation reactor technology.
Why Nuclear? The AI Power Problem
AI training and inference are not intermittent workloads. A cluster of 100,000 GPUs cannot pause while the sun goes down or the wind stops blowing. It needs power 24 hours a day, 365 days a year, with no interruptions. That's what nuclear provides.
Wind and solar cannot meet the "five-nines" (99.999%) uptime requirements. Those are the demands of modern AI training and inference, as Wedbush analysts noted. Nuclear provides "constant-on" power. It avoids the intermittency of solar and wind. Those sources need massive battery storage to keep data centers stable.
One senior researcher from the Pacific Northwest National Laboratory captured the shift: "We are seeing a shift from 'AI at any cost' to 'AI at any wattage'."
The cost calculus has changed. A single data center campus can require a gigawatt of power. An AI model training run can consume as much electricity as a small city. When that is true, power reliability matters as much as chip performance.
The Risks: What Could Go Wrong
Tech companies are the least risk-tolerant organizations on the planet. They spend billions on security, redundancy, and failover. Nuclear power is the most heavily regulated industry in the world. Construction timelines run decades. Cost overruns run billions.
The clash between these two worlds is where the story gets interesting.
Timing. Constellation is on track for a 2027 restart. But grid operator PJM says the plant may not connect until 2031. The reactor will be ready. The transmission infrastructure will not. Constellation has asked federal regulators for help speeding up the process. A decision could come as early as June or July, but the timeline remains uncertain.
The engineering challenge. Constellation is spending $1.6 billion to refurbish a 50-year-old reactor. The plant's analog control systems are being retained because they provide an "air-gapped" security advantage against sophisticated cyberattacks. This is a bet that old technology can meet new demands.
Regulatory uncertainty. Amazon's behind-the-meter strategy at Susquehanna faced major legal challenges. They came at the Federal Energy Regulatory Commission in 2024 and 2025. Critics argued that co-located data centers "free-ride" on the grid. A pivotal 5th U.S. Circuit Court ruling and new FERC rulemaking have cleared a path, but the regulatory environment remains unsettled.
Cost overruns. The $1.6 billion refurbishment cost for Three Mile Island is just the beginning. The full build-out will cost tens of billions of dollars. That includes Amazon's 5 GW SMR program, Meta's 6.6 GW portfolio, and Google's 500 MW Kairos fleet. When the hyperscalers are "willing to take on the risks of cost overruns and delays," as one portfolio manager put it, that determines "just how much of a boost these agreements give the sector."
Public perception. Three Mile Island is the site of America's worst commercial nuclear accident. Environmental groups have expressed caution regarding the restart. The plant's name is being changed to the Crane Clean Energy Center for a reason.
What This Means
The nuclear play is not about idealism. It's not about climate change. It's about power — literally. AI needs more electricity than the grid can provide. The grid cannot be rebuilt fast enough. So the companies that need the power are building their own.
This is a structural shift in the relationship between technology and energy. For decades, tech companies consumed electricity. They optimized for efficiency. They bought renewable energy credits. But they didn't build power plants. That era is ending. Microsoft is restarting a nuclear reactor. Amazon is building data centers inside nuclear facilities. Google is funding advanced reactor technology. Meta is underwriting 6.6 gigawatts of new nuclear capacity.
AI doesn't care about politics. It cares about electrons. And the only way to get enough of them is to go nuclear. That is the scale and reliability required.
Sources: Wedbush analyst reports (January 2026); Reuters (March 26, 2026; April 10, 2026); AP News (January 9, 2026); UPI (January 9, 2026); 9to5Google (June 15, 2026); Nuclear Engineering International (April 21, 2026); World-Energy.org (March 27, 2026); Kairos Power official announcements (April–June 2026); NRC Crane Clean Energy Center documentation. The
Disclaimer
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations
This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.
Forecasts from third-party analysts can change with market conditions.
Cost and pricing examples are point-in-time estimates; actual rates vary.
Country and company comparisons rely on public reporting, not operational data.
This sector moves fast; timelines and deal terms may be updated later.
Company deals and regulatory rulings may evolve; verify current status.
AI infrastructure is changing quickly; claims can become outdated soon.
Sources
- Wedbush analyst reports (January 2026)
- Reuters (March 26, 2026; April 10, 2026)
- AP News (January 9, 2026)
- UPI (January 9, 2026)
- 9to5Google (June 15, 2026)
- Nuclear Engineering International (April 21, 2026)
- World-Energy.org (March 27, 2026)
- Kairos Power official announcements (April–June 2026)
- NRC Crane Clean Energy Center documentation.
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations: This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.; Forecasts from third-party analysts can change with market conditions.; Cost and pricing examples are point-in-time estimates; actual rates vary.; Country and company comparisons rely on public reporting, not operational data.; This sector moves fast; timelines and deal terms may be updated later.; Company deals and regulatory rulings may evolve; verify current status.; AI infrastructure is changing quickly; claims can become outdated soon.