A 12-gigawatt, twenty-year nuclear power purchase agreement signed late Friday between one of the country’s largest cloud providers and a consortium led by Constellation Energy and the small-modular reactor developer X-energy has reset expectations for how the artificial intelligence build-out will be powered, with energy executives, grid operators and Wall Street analysts spending the weekend recasting models that had assumed natural gas would carry most of the new load.

The agreement, the largest single corporate nuclear procurement on record, covers output from three existing reactors slated for life extension and from a fleet of small-modular units to be built across four sites in Pennsylvania, Maryland and Texas between 2028 and 2034. People briefed on the deal, which has not been publicly disclosed in full, said the hyperscaler had agreed to a blended take-or-pay price in the high-$70s per megawatt-hour, with a floor and a ceiling tied to a basket of capacity-market and inflation indices.

The structure, which guarantees a long-dated revenue stream to the consortium and locks in capacity for the buyer, was described by two energy bankers as “the template the rest of the industry was waiting for.” Several other hyperscaler procurement teams are now expected to accelerate similar pacts, the bankers said, with at least three deals already in advanced negotiation.

“This is the deal that proves you can finance an SMR on a corporate signature, not a state subsidy,” said Ramona Kettlewell, head of power and utilities research at the investment bank Trent Harlow. “Until Friday, every SMR project in this country was waiting on either federal loan guarantees or a state procurement vehicle. Friday changed the math.”

The buyer, identified by four people familiar with the negotiations but which declined to be named publicly until the deal’s terms are filed with state regulators in the coming week, has guided to roughly $58 billion in 2026 capital expenditure, the majority directed at AI training infrastructure. The 12 gigawatts contracted on Friday is sufficient, on an annual basis, to power approximately ten of the company’s largest planned training campuses, according to grid-load filings reviewed by analysts at the consultancy Rhodium Power.

The agreement lands in a markedly different political climate than even six weeks ago. The collapse of the Sanders-Ocasio-Cortez federal training moratorium in the House Ways and Means Committee on April 22 lifted the most acute regulatory overhang from new construction. But state-level moratorium bills in California, New York and Washington — and a parallel disclosure measure unveiled in the Senate on Wednesday by a bipartisan group led by Sens. Maggie Hennessey, D-Colo., and Travis Beaumont, R-Ohio — have kept pressure on hyperscalers to demonstrate that the build-out will not strain household electricity bills or push grid operators into a fresh round of gas-fired procurement.

“The moratorium fight made clear that the political license to build was conditioned on a credible answer to the grid question,” said Davis Okonkwo, a senior fellow at the Energy Futures Initiative. “Nuclear is a credible answer. It is not the only one, and it is not a fast one, but a 12-gigawatt signature gets you out of the worst version of the conversation.”

Grid operators reacted cautiously. PJM Interconnection, which manages the wholesale market that covers most of the contracted capacity, said in a brief statement Saturday that it had been informed of the agreement in general terms but that “specific interconnection and queue implications will be determined through our standard process.” A spokesperson for the Electric Reliability Council of Texas declined to comment on the Texas portion of the deal.

Critics of the AI build-out argued the announcement, while welcome on emissions grounds, did little to address the near-term load growth that has already pushed several utilities to defer coal-plant retirements. The first kilowatt-hour from the new SMRs is not expected to flow before 2029.

“Between now and the end of the decade, you are still talking about gas and, in some places, coal staying on the system longer than the climate plans assumed,” said Mira Saldanha, a grid analyst at the advocacy group Public Power Project. “A 2034 reactor does not run a 2027 data center. The honest version of this story is that nuclear is part of the answer for the back half of this build-out, and ratepayers are part of the answer for the front half.”

Constellation, which would supply roughly 7.5 gigawatts of the contracted volume from its existing fleet and from uprates, declined through a spokesperson to comment on specific commercial terms but confirmed the agreement in general. X-energy, which will supply the small-modular portion under a phased deployment, said in a written statement that the deal would “underwrite the largest single advanced-reactor order ever placed in the United States” and accelerate its supply-chain plans in Tennessee and South Carolina.

Equity analysts have begun adjusting price targets across the nuclear-adjacent complex. Two pure-play SMR developers traded sharply higher in pre-market activity in Asia on Sunday on news of the agreement, according to traders at Asian desks, with U.S. trading set to open Monday. A note circulated Sunday afternoon by the brokerage Linhart & Co. raised earnings estimates for the consortium’s lead utility by an average of 9 percent over the next three fiscal years.

The deal has also drawn the attention of regulators outside the energy domain. The Federal Trade Commission, which has been examining concentration in the AI labor market, indicated through a spokesperson Sunday that “long-term exclusive power contracts” between hyperscalers and generators were among the structures it had been studying as part of a broader competition review. The spokesperson declined to comment on whether the Friday agreement was a specific subject of inquiry.

For the buyer, the contracting cost is meaningful but modest against the scale of the AI investment cycle. At the blended price reported, the deal commits roughly $7.5 billion to $8.5 billion a year in power purchases at full ramp — a number senior executives at peer firms have privately described as “the price of being allowed to keep building,” according to a person who attended a closed-door industry session in Washington earlier this month.

“The capex story for AI has always had a denominator nobody wanted to talk about, which is gigawatts,” said Kettlewell, the analyst. “On Friday somebody finally wrote a check for the denominator. That is going to focus a lot of minds in boardrooms this week.”

Industry observers said additional power-procurement announcements were likely from rival hyperscalers in the coming weeks, with several boards expected to take up similar proposals at meetings scheduled before the end of the month.