The United States now has 74 gigawatts of planned or new nuclear reactor capacity, driven largely by electricity demand from AI data centers, according to a new analysis published by the Breakthrough Institute. While it's unlikely all of that capacity will ultimately be built, even a fraction deployed would mark a historic turnaround for an industry that has struggled for decades. The report finds the profile of nuclear buyers has fundamentally changed, with hyperscale data center companies now dominating the commercial pipeline.

Companies building hyperscale data centers are responsible for 30 gigawatts, or 41%, of the 74 gigawatts of announced or prospective new nuclear reactor capacity in the United States. That's more than double the 18% currently under order from utilities and public-power buyers. If utility-hyperscaler hybrid deals are included, the total share of the commercial pipeline drawn from hyperscaler demand is nearly half. The report's dataset groups the commercial pipeline into three maturity stages—building, firm, and tentative—across five specific tiers. Nearly 60 gigawatts of announced or prospective capacity sits in fleet-level demand signals in Tier 5, the least mature stage, while more mature tiers remain comparatively small. Hyperscalers have a direct stake in 45% of the roughly 60 gigawatts in Tier 5, with hyperscaler-only entries accounting for about 21 gigawatts, or 35%, and utility-hyperscaler hybrid deals adding another 6 gigawatts, or 10%.

The report finds that hyperscalers are already showing up in the existing nuclear fleet, where capacity can be delivered faster. Hyperscalers are "the clear catalyst" behind 8.3 gigawatts of existing-fleet nuclear capacity: roughly 6.3 gigawatts through existing-plant power purchase agreements, about 1.5 gigawatts through restart-linked agreements, and about 0.5 gigawatts through uprate-linked deals. According to the analysis, the demand pipeline isn't simply a return to large light-water reactor deployment. Across all stages and tiers except for Tier 2, small light-water reactors and non-light-water reactors account for most announced or prospective capacity, even though large reactor units are much bigger individually. The report states this suggests "the pipeline is far more technologically diverse than the historical U.S. buildout."

The report explains that hyperscalers and data center developers are looking for large amounts of reliable, clean electricity, and nuclear is one of the few technologies that can plausibly provide firm power at that scale. The authors write that these early shifts in reactor demand should be understood as evidence that the U.S. is in a "market formation process: the demand has arrived well ahead of the technological products, regulatory certainty, supply chain maturity, and financing support." Commercial nuclear demand is shifting away from decades of exclusive dependence on utilities and public-power buyers, but this transition is still in its early stages. The reality is more complicated than AI demand simply rescuing nuclear: non-utility buyers are creating much of the visible demand pressure, while utilities and public institutions still provide much of the deployment interface.

The report concludes that the more likely future is a "matrix of orderbooks"—repeatable project structures matched to different use cases. Large reactors may be best suited to bulk power and regional load growth, small modular reactors may fit public power systems and coal-site replacement, and microreactors may scale first through defense and remote markets. The point, according to the analysis, isn't to pick a single winner but to create repeatable combinations of reactor designs, customers, sites, financing structures, and licensing pathways. For the first time in decades, new nuclear is being ordered by large, fast-growing customers that need firm, clean, around-the-clock power at scale, and the sector is no longer waiting only on traditional utility procurement.