Big Tech is signing its first direct deals to buy nuclear power, reshaping how AI gets its electricity

Companies that run the largest data centers on the planet are no longer content to simply buy electricity off the regional grid and hope enough of it stays reliable. A handful of the biggest names in cloud computing and artificial intelligence have begun signing direct agreements with nuclear power developers, locking in electricity from reactors that in some cases have not yet been built. The shift marks one of the more unusual pairings in modern industry: a technology sector built on rapid iteration and quarterly product cycles betting on an energy source known for multi-decade construction timelines and heavy regulation.

Data centers’ expanding appetite for electricity

The scale of the problem driving these deals is enormous. Electricity demand from data centers is projected to roughly double by 2027, with artificial intelligence workloads responsible for more than 60% of that growth. Training and running large AI models requires banks of specialized processors that run continuously at near-full capacity, a load profile very different from the more variable draw of a typical office building or factory. Utilities in several regions have already warned that new data center campuses could outpace planned grid upgrades, pushing technology firms to look for power sources they can control directly rather than wait in line behind other customers. In some markets, a single proposed data center campus now draws as much power as a mid-sized city, a scale of demand that grid planners historically spread across years of incremental industrial and residential growth rather than a handful of corporate customers signing contracts within the same budget cycle. That mismatch between how fast a data center can be built and how slowly new transmission lines and power plants typically get approved is the underlying force pushing technology firms toward unconventional supply arrangements.

Grid operators have responded by asking large prospective customers to disclose their power needs years in advance, a departure from the standard practice of connecting new customers as capacity becomes available. That shift in posture reflects how disruptive an unplanned data center campus can be to a regional grid that was sized around more predictable, slower-growing demand.

Why nuclear power fits AI’s constant demand

Nuclear reactors produce electricity around the clock regardless of weather, which makes them attractive to operators whose servers cannot tolerate the intermittent output of solar or wind without extensive battery backup. The Department of Energy has noted that pairing a data center with a dedicated or co-located nuclear plant can provide the steady, carbon-free baseload power that AI training clusters need, while also reducing the strain that a sudden new multi-hundred-megawatt customer would otherwise place on a shared transmission network. That combination of reliability and a fixed, contracted price has turned nuclear from a niche curiosity in corporate energy planning into a serious line item for firms that once relied almost exclusively on wind and solar purchase agreements to hit their sustainability targets.

TerraPower and the new wave of power-purchase agreements

Some of the earliest concrete moves have come from advanced reactor developers such as TerraPower, whose next-generation design has become a reference point for how these arrangements might work in practice. Rather than waiting for a plant to reach commercial operation before negotiating a sale, technology buyers are signing power-purchase agreements years ahead of a reactor’s expected completion date, effectively helping to finance construction in exchange for a guaranteed future electricity supply. That structure gives reactor developers the long-term revenue certainty they need to secure financing and regulatory approval, while giving data center operators a contracted claim on carbon-free power before a competitor can lock up the same capacity.

Small modular reactors versus existing plants

Two distinct strategies have emerged among technology buyers. Some have pursued agreements tied to small modular reactors, compact designs that can theoretically be built faster and sited closer to a data center campus than a traditional large reactor, though none of the newest designs has yet completed construction at commercial scale in the United States. Others have gone after capacity from existing nuclear plants already connected to the grid, in some cases supporting the restart of previously shuttered reactors or the license extension of aging ones. The two approaches carry different risk profiles: betting on a next-generation design means accepting construction and regulatory uncertainty in exchange for a purpose-built supply, while buying from an operating plant offers power sooner but limits how much new capacity actually gets added to the grid. A third, less publicized option involves technology companies taking minority financial stakes in reactor projects directly, an arrangement that goes beyond a simple purchase agreement and ties the buyer’s balance sheet more closely to whether a given reactor design succeeds. That deeper financial entanglement is itself a signal of how seriously some buyers now treat electricity supply as a strategic risk on par with chip supply or data center real estate, rather than a routine utility expense handled by a facilities department.

Regulatory and grid hurdles ahead

None of these arrangements bypasses the lengthy approval process that has defined nuclear construction in the United States for decades. New reactor designs still require licensing review, and connecting any large new power source to the grid involves interconnection studies that can themselves take years to complete. The Department of Energy’s own assessment of nuclear-powered data centers points to financing, permitting timelines, and workforce availability as the central obstacles standing between today’s signed agreements and reactors actually delivering electricity. How quickly those hurdles clear will determine whether direct nuclear deals become a mainstream tool for powering artificial intelligence or remain a small, high-profile slice of a much larger and still mostly fossil-fuel-and-renewables-dependent energy mix.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

More from Morning Overview