Google Locks In 3.6 Gigawatts From Constellation as AI Power Race Intensifies

Google has signed one of the largest power deals in its history, agreeing to purchase roughly 3,590 megawatts of electricity from Constellation Energy in a package that funds capacity upgrades at 11 nuclear reactors across Illinois, Pennsylvania, and New Jersey. The agreement, confirmed by both companies this week, unlocks more than $4.3 billion in new Constellation investment and underscores how securing grid capacity, not chip supply, has become the binding constraint on AI data center expansion.

Of the total capacity, 890 megawatts will come from uprates at existing nuclear units — plant upgrades that squeeze more output from reactors already on the grid rather than building new generation from scratch. The remaining roughly 2,700 megawatts is a 15-year supply agreement spanning the PJM Interconnection territory without being tied to a specific plant. Google says the first uprated capacity could reach the grid as early as 2028.

Why PJM, and why now

The deal lands squarely in PJM Interconnection, the 13-state grid operator serving 67 million people from Illinois to Virginia, which has become ground zero for the collision between AI data center growth and aging grid infrastructure. PJM has pushed proposals requiring large data center customers to arrange their own backup power or risk being disconnected during peak demand, a response to fears that unpredictable AI compute loads could destabilize a grid already straining to keep pace with new construction.

Our agreement with Constellation to fund nuclear reactor uprates will strengthen the PJM grid, which serves 67 million people.

That line from Amanda Peterson Corio, Google's global head of energy and power, frames the deal less as a clean-energy gesture and more as grid insurance — buying firm, always-on nuclear capacity that can absorb data center load swings in a way that intermittent wind or solar cannot. Constellation CEO Joe Dominguez struck a similar note, calling the collaboration "a model for how technology companies and the energy industry can work together," according to Data Center Dynamics' report on the agreement.

A pattern across the industry

Google's move follows a now-familiar sequence among hyperscalers. Meta has committed to roughly 6.6 gigawatts from nuclear developers including Vistra, TerraPower, and Oklo, plus 5.2 gigawatts of gas-and-battery capacity through Entergy Louisiana. Microsoft has locked in 2.67 gigawatts of natural gas generation, and Amazon has added 690 megawatts of nuclear supply of its own. Collectively, these deals signal that frontier AI buildouts are now being paced by multi-year utility contracts and reactor licensing timelines rather than GPU shipment schedules.

The scale of that shift shows up in national demand figures: total U.S. electricity consumption is projected to climb from 4.195 trillion kilowatt-hours to 4.356 trillion kilowatt-hours by 2027, a jump utilities attribute substantially to data center growth. Constellation says the Google agreement will also preserve roughly 4,400 existing jobs at its nuclear plants and create about 7,200 positions during the uprate construction work.

The political backdrop

The timing is not incidental. The deal follows pressure from the Trump administration for large technology companies to shoulder the cost of new power infrastructure directly rather than passing it through to residential ratepayers, after a wave of public backlash over rising utility bills near major data center clusters. Structuring deals as long-term bilateral PPAs — rather than relying on the general PJM capacity market — lets Google argue its new demand is funding its own supply, even as critics note these contracts still compete for scarce interconnection queue slots and skilled nuclear labor that the broader grid also needs.

For supply-chain vendors serving the sector, the takeaway is that nuclear uprate work, grid interconnection equipment, and long-cycle utility contracts are becoming as central to AI infrastructure planning as silicon itself — and Constellation, with multiple hyperscaler deals now on its books, has emerged as one of the AI boom's primary power suppliers.