A Nuclear Plant Running Since 1986 Is About to Get 348 Megawatts Bigger. Google Is Paying the $4.3 Billion Bill, Not Ratepayers.

Limerick Generating Station has been running along the Schuylkill River in Montgomery County, Pennsylvania, since 1986. Nobody is building a new reactor there, and nobody is proposing one. But by the early 2030s, the plant's two units will put out 348 more megawatts than they do today, and the money paying for it is coming from Google under a 20-year contract signed October 6th, not from anyone's electric bill.

The deal is bigger than Limerick alone. Constellation Energy and Google announced a 20-year power purchase agreement covering 890 megawatts of new nuclear capacity, all of it from uprates — government-approved upgrades that let an existing reactor produce more electricity than its original license allowed — spread across 11 units Constellation operates inside PJM, the grid operator covering Pennsylvania, New Jersey, Maryland, Ohio, and most of the mid-Atlantic and parts of the Midwest. Limerick's two units carry the single largest share, 348 MW. Salem Generating Station, in southern New Jersey, adds another 97 MW. Constellation puts the total cost of the program at $4.3 billion.

An uprate is not a new power plant

It's a set of equipment changes — bigger turbines, upgraded steam generators and condensers, recalibrated instrumentation — that squeeze more output out of a reactor that's already built, already staffed, and already connected to the grid. Each change still needs its own license amendment from the Nuclear Regulatory Commission, and Constellation has run that process before: an earlier uprate program at its Byron and Braidwood plants in Illinois added roughly 135 megawatts for about $800 million. The advantage over building new nuclear is almost entirely about time. A reactor uprate can clear licensing and construction in a few years. A new reactor, even a small one, routinely takes a decade or more once permitting is included.

I think the 890-megawatt headline number undersells what's actually happening here. The hard, expensive, multi-year part of a nuclear project — siting, containment, cooling systems, transmission interconnection — is already built and already licensed at all 11 of these units. What's left is equipment and paperwork, and Google just agreed to pay for both.

The ratepayer math, up front

Google and Constellation built the announcement around the White House's Ratepayer Protection Pledge, the federal effort aimed at keeping AI companies' power demand off ordinary electric bills. The structure is direct: Google signs a 20-year contract to buy the new output, Constellation uses that contract to finance the $4.3 billion build-out, and PJM gets 890 MW of new, firm, carbon-free capacity that didn't exist before — without any utility asking a state commission to fold construction costs into anyone's base rate.

Preserving and expanding the output of the existing nuclear fleet and developing new clean generation is critical to meeting our nation's growing energy needs while keeping costs affordable for homes and businesses.Joe Dominguez, Chairman, President, and CEO, Constellation Energy

The timing matters. PJM's capacity market has been stretched: its September backstop auction came up 6.8 gigawatts short of the grid operator's own reliability target, with Monitoring Analytics, PJM's independent market monitor, naming data-center load as the single biggest driver. An 890 MW addition phased in between 2028 and 2032 doesn't close a gap that size by itself, but it's real new supply landing inside the exact footprint where the shortage is worst, financed by the company adding the new demand rather than by the people already on the grid.

A different kind of hyperscaler nuclear deal

This is the second Constellation-hyperscaler nuclear agreement in two weeks. Amazon signed a 20-year PPA for 690 MW of Calvert Cliffs' existing output the day after FERC froze PJM's own backstop auction — a deal for power the plant already makes. Google's agreement works the other direction: instead of buying existing capacity, it funds new capacity at 11 plants, with NRC license amendments still to come at each one before any of the 890 MW is real. On GridPermit's evidence ladder, a signed 20-year PPA with a named capital figure and a per-plant megawatt table climbs higher than a round corporate pledge to decarbonize — but those license amendments are the next rung, not yet climbed, and worth checking against the NRC's own docket as each one gets filed.

Most of this year's data-center power coverage has been about building generation from nothing: new gas turbines, new reactors, grid-scale batteries, all with long lead times and plenty of ways to slip. An uprate skips most of that. The containment building, the cooling system, and the transmission interconnection are already standing and already licensed. If utilities and their biggest customers are looking for the fastest real megawatts in PJM, the plants already running may be the cheapest place to find them.