Meta has committed to three nuclear power deals that could deliver up to 6.6 gigawatts of electricity for its artificial intelligence data centers by 2035, a volume of contracted capacity that dwarfs any prior single-company nuclear procurement in the United States. The agreements, struck in 2026, tie the social media giant’s AI expansion directly to the output of nuclear plants at a time when the U.S. Nuclear Regulatory Commission already faces growing pressure from power uprate applications. The scale of this commitment raises a pointed question: can the federal licensing process keep pace with the speed at which hyperscale tech companies are locking up generation capacity?
Why Meta’s 6.6-gigawatt nuclear bet matters now
The sheer size of Meta’s nuclear procurement, up to 6.6 GW by 2035 according to the Associated Press, is roughly equivalent to the output of six to seven large commercial reactors running at full power. That capacity is not being built from scratch overnight. A significant share of it depends on squeezing more megawatts from existing plants through a regulatory process known as a power uprate, which requires the NRC to review and approve changes to a reactor’s licensed thermal output before any additional electricity can flow to the grid.
Power uprates come in three categories, from small measurement-uncertainty recaptures to extended uprates that can boost output by up to 20 percent. Each type demands a formal license amendment, safety analysis, and NRC staff review. Vistra, one of the largest U.S. nuclear operators, has claimed 433 MW of uprates across its fleet, according to the NRC’s published power uprate program. If multiple operators file simultaneous uprate applications to serve hyperscale AI loads, the review queue could grow faster than the agency’s technical staff can process it.
The practical risk is straightforward. When a single plant like Davis-Besse files for an uprate, the NRC reviews the application, its supplements, and supporting safety analyses on a defined timeline. Stack several such filings at once, driven by contracts with companies like Meta that need power on aggressive schedules, and average review times could stretch well beyond historical norms. A 30 percent increase in review duration is a reasonable estimate given the agency’s staffing constraints and the complexity of simultaneous safety evaluations, though the NRC has not published a formal projection on this point.
Contracts, uprates, and the Davis-Besse precedent
The clearest window into how nuclear capacity actually gets unlocked for new customers is the NRC’s own licensing record. The agency maintains a detailed public file for the Davis-Besse uprate, including ADAMS accession numbers for the original application, technical supplements, and the final license amendment. That paper trail shows how granular and time-consuming the process is: each filing requires engineering justification, independent safety review, and a formal NRC decision before a single additional megawatt can be generated.
Davis-Besse’s case involved a measurement uncertainty recapture uprate, the smallest and typically fastest category. Even so, the documentation runs to dozens of filings and supplements. Extended power uprates, which deliver larger capacity gains, take longer and demand more extensive plant modifications, from upgraded pumps and turbines to enhanced safety systems. If Meta’s 6.6 GW target depends on a mix of uprate types across multiple plants, the cumulative regulatory workload will be substantial and highly sensitive to bottlenecks inside the NRC’s review offices.
Meta itself has stated that the three 2026 deals support up to 6.6 GW by 2035. That language, “up to,” signals a ceiling rather than a guarantee. The actual delivered capacity will depend on how quickly plant operators can secure NRC approvals, complete any required hardware upgrades, and bring additional output online. Each step carries its own timeline risk, and none of it is under Meta’s direct control. A delay in a single plant’s uprate could ripple through Meta’s data center buildout if that capacity was earmarked for a specific AI cluster or regional hub.
The 433 MW of uprates claimed by Vistra, while significant for a single operator, represents a fraction of what Meta’s full 6.6 GW ambition would require. Reaching that target likely means tapping new reactor construction, long-term power purchase agreements with existing plants running at current capacity, or some combination of both, alongside uprates. The public record does not yet show the full set of plants or operators tied to Meta’s contracts, leaving a gap between the headline number and the documented capacity pipeline. That opacity makes it difficult for outside observers to map Meta’s AI growth plans onto specific nuclear assets or licensing milestones.
Unanswered questions about Meta’s nuclear timeline
Several critical pieces of this story remain unresolved. The complete ADAMS filings and technical safety analyses for every reactor tied to Meta’s contracts have not been publicly identified beyond the single Davis-Besse example. Without that documentation, it is not possible to assess how much of the 6.6 GW depends on uprates versus new construction versus existing uncommitted capacity. Each pathway carries different risks: uprates hinge on regulatory throughput, new-build reactors face construction and financing hurdles, and existing capacity can be constrained by competing regional demand.
The official NRC approval status and projected timelines for the full set of uprates referenced in the 2026 deals are also absent from the public record. Primary contract documents or capacity schedules that specify exact megawatts Meta has committed to purchase from each plant have not been disclosed. That means the gap between Meta’s stated ambition and the concrete, licensed capacity supporting it is still largely a matter of inference.
For regulators, this lack of transparency complicates planning. If multiple utilities are quietly aligning their uprate strategies around a handful of hyperscale customers, the NRC could face a surge of complex applications without a clear public signal of what is driving the spike. For grid planners and state regulators, the opacity makes it harder to judge whether regional power systems will be able to support both AI-driven demand and traditional loads without stressing transmission or reliability margins.
For Meta and its peers, the unanswered questions translate into execution risk. AI data centers are capital-intensive, and their business models assume reliable, long-term access to low-carbon electricity. If uprate approvals slip, if construction schedules on new reactors extend, or if existing nuclear plants face unexpected outages, the carefully balanced timelines that underlie the 6.6 GW target could unravel. That would not only affect Meta’s internal road maps but could also reshape how investors and policymakers view the role of nuclear in supporting digital infrastructure.
What is clear from the available record is that the NRC’s licensing machinery, designed for a slower and more predictable era of nuclear expansion, is now being tested by the tempo of AI-era demand. Whether that machinery can adapt-through staffing, process reforms, or clearer coordination with major power buyers-will go a long way toward determining how much of Meta’s nuclear vision becomes reality by 2035, and how quickly other tech giants can follow the same path.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
