Earlier this year, Meta made one of the largest corporate bets on nuclear power in American history, announcing three deals that could deliver up to 6.6 gigawatts of atomic energy to feed its growing fleet of artificial intelligence data centers by 2035. The centerpiece is a pair of 20-year power purchase agreements with Vistra Corp. covering more than 2,600 megawatts of nuclear capacity at plants in Ohio and Pennsylvania, enough electricity to power roughly two million homes.
The announcements, made in January 2026, landed at a moment when the biggest names in tech are scrambling to lock down reliable, carbon-free power for AI workloads that run around the clock and consume electricity at industrial scale. Microsoft signed a deal to restart a unit at Three Mile Island. Amazon purchased a nuclear-powered data center campus in Pennsylvania. Google invested in geothermal startups. Meta’s move dwarfs them all in raw megawatts and signals that the company sees nuclear energy not as a niche experiment but as the backbone of its AI infrastructure.
The Vistra deal: 2,600 megawatts locked in
The most concrete piece of Meta’s nuclear strategy is the agreement with Vistra, one of the largest power producers in the United States. According to a Vistra regulatory disclosure, the 20-year contracts cover 2,176 megawatts of existing operating nuclear capacity plus 433 megawatts of planned output increases, known as uprates, at three facilities: Perry and Davis-Besse in Ohio and Beaver Valley in Pennsylvania.
All three plants sit within the PJM Interconnection, the regional grid operator that manages electricity delivery across 13 states and the District of Columbia. That geography matters. PJM territory, stretching from Virginia through Ohio, has become the epicenter of data center construction in the United States, and the queue for new grid connections has ballooned to years-long waits. Because the Vistra plants are already built and operating, Meta sidesteps much of that bottleneck and gains immediate access to baseload power that runs day and night regardless of weather.
Meta has tied the nuclear supply to its data center campus in New Albany, Ohio, which the company has described as a long-term hub for training and running large AI models. Training a single frontier AI model can consume as much electricity as tens of thousands of households use in a year, and the demand only grows as models get larger. Solar and wind farms can contribute, but their output fluctuates with sunlight and weather. Nuclear plants, by contrast, typically run at more than 90 percent capacity around the clock, making them a natural fit for workloads that cannot tolerate interruptions.
The next-generation gamble: Oklo and TerraPower
The remaining roughly 4,000 megawatts needed to reach Meta’s 6.6-gigawatt target depend on reactor technologies that have never operated commercially. Two companies figure prominently in that ambition.
Oklo, a startup developing compact fast reactors, is working toward operation of its first Aurora unit by 2027. Fuel preparation is underway at Idaho National Laboratory’s fuel fabrication facility, and the company has attracted significant venture capital and a public listing. But the 2027 target refers to a single demonstration unit, not a fleet of reactors sized for data center demand. No publicly available filing confirms a binding power purchase agreement between Oklo and Meta.
TerraPower, the Bill Gates-backed nuclear venture, is building its Natrium demonstration reactor in Kemmerer, Wyoming, under a cooperative agreement with the U.S. Department of Energy. The sodium-cooled design promises flexible output and built-in energy storage, features that could complement data center loads. But like Oklo’s Aurora, Natrium remains a demonstration project. Its timeline for commercial electricity delivery has not been publicly fixed, and no confirmed offtake contract with Meta has surfaced in federal filings.
Both projects carry the kinds of risk that come with first-of-a-kind engineering: regulatory reviews that stretch longer than planned, design changes uncovered during construction, and local permitting battles that can stall timelines. History offers a cautionary note. The Vogtle nuclear expansion in Georgia, using a more conventional reactor design, ran roughly seven years behind schedule and billions of dollars over budget before its two new units finally came online.
What the ‘up to’ qualifier really means
Meta’s framing of the deals as supporting “up to 6.6 gigawatts by 2035” deserves careful reading. According to Associated Press reporting, that figure aggregates the confirmed Vistra capacity with the aspirational output from next-generation projects. The phrase “up to” is doing significant work: it means the final delivered capacity could land well below that ceiling depending on construction progress, licensing outcomes, and reactor performance.
For scale, 6.6 gigawatts is roughly the output of six or seven large conventional nuclear plants. Reaching that level by 2035 would require not just the Vistra plants running at full capacity but the successful construction, licensing, and commercial operation of reactor designs that exist today only as prototypes or engineering models. It would also require timely completion of grid upgrades and interconnection work in a region where utilities and regulators are already struggling to keep pace with electricity demand.
The gap between the roughly 2,600 megawatts backed by detailed Vistra contracts and the 6.6-gigawatt headline number illustrates a pattern common across Big Tech energy announcements: companies publicize their maximum ambition while the binding commitments cover a fraction of the total. That does not make the ambition hollow. It does mean that readers, investors, and policymakers should treat the larger figure as a strategic signal about Meta’s preferred energy direction rather than a fully contracted buildout plan.
Unanswered questions for ratepayers and regulators
Several important details remain undisclosed. Vistra and Meta have not revealed the financial terms of the 20-year contracts, including how power prices will adjust over time, how costs will be allocated if uprates are delayed, or whether the agreements include provisions for carbon pricing changes. Long-duration nuclear contracts can involve complex hedging structures, and without visibility into those provisions, it is difficult to assess how much economic risk Meta is absorbing versus passing along.
The 433 megawatts of planned uprates at the three Vistra plants also require engineering work and Nuclear Regulatory Commission review before they can deliver additional power. No public interconnection study or cost estimate for those uprates has been released. Consumer advocates and grid watchdogs will want to know whether the uprate costs will be borne entirely by Meta and Vistra or whether any portion could flow through to ratepayers served by PJM utilities.
There is also the broader question of what happens to electricity markets when a single corporate buyer locks up thousands of megawatts of baseload generation for two decades. Critics of similar deals, including the Microsoft-Constellation arrangement to restart Three Mile Island Unit 1, have argued that diverting nuclear output to private buyers could tighten supply for everyone else and push wholesale electricity prices higher. Meta and Vistra have not publicly addressed how their agreements interact with PJM’s capacity market or whether the contracts include provisions to release power back to the grid during emergencies.
Why nuclear, and why now
The timing of Meta’s nuclear push reflects a collision of two forces. On one side, AI model training and inference are consuming electricity at a pace that has caught even utility planners off guard. PJM projected in late 2025 that data center electricity demand in its territory could more than double by the early 2030s. On the other side, the existing U.S. nuclear fleet, which generates roughly 20 percent of the nation’s electricity, faces an uncertain economic future as plants age and cheaper natural gas and renewables compete for market share.
Deals like Meta’s offer nuclear plant owners a lifeline: guaranteed revenue over decades, which justifies the investment needed to keep aging reactors running and to squeeze additional output from them through uprates. For Meta, the appeal is equally straightforward. Nuclear provides the kind of firm, 24/7, zero-carbon electricity that no combination of solar panels and battery storage can yet match at data center scale, at least not without enormous land footprints and grid storage investments that remain prohibitively expensive.
Whether Meta’s bet pays off in full depends on variables that no contract can control: the pace of advanced reactor development, the willingness of regulators to license new designs, the stability of federal energy policy, and the continued growth of AI workloads that justify this level of infrastructure spending. What is already clear, based on the Vistra agreements alone, is that nuclear power has moved from the margins of the tech industry’s energy strategy to its center. The remaining question is how much of Meta’s 6.6-gigawatt vision will materialize as operating reactors and how much will remain ambition on paper.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
