Google Will Pay $4.3 Billion to Upgrade Existing Reactors for 890 Megawatts -- Its Fastest Nuclear Deal Yet

Google and Constellation Energy said Tuesday they'll spend more than $4.3 billion upgrading 11 of Constellation's nuclear reactors across Illinois, Pennsylvania and New Jersey, adding 890 megawatts of new capacity to PJM, the grid serving roughly 65 million people across 13 Eastern states and the District of Columbia. A 20-year power purchase agreement covers that new output; a separate 15-year agreement locks in 2,700 MW Constellation already generates today. The first uprated capacity is expected online by 2028 -- fast, by nuclear standards, because the plants in question already exist and are already running.

Wall Street reacted immediately: Constellation shares jumped 12% the morning the deal was announced, with merchant nuclear peers Vistra and Talen Energy climbing 8% and 7% on the same news despite having no stated role in it -- a read-through rally on the bet that more hyperscalers will pay utilities to upgrade reactors rather than build new ones. Constellation stock remains down for the year even after the jump, a reminder that the rally reflects this deal's structure more than a reversal in sentiment about nuclear economics broadly.

The deal is Google's fifth major nuclear commitment since October 2024, and each one measures something different -- a small modular reactor fleet that doesn't exist yet, early-stage development capital for sites still being sited, a restart of a plant shuttered since 2020, a share of an overseas plant's existing output, and now an uprade of reactors that have been online for decades.

Lined up, the five deals show two very different timelines wearing the same "nuclear for AI" label. The small-reactor and restart bets require building new technology or reviving a plant shut for years; NextEra isn't targeting power from Duane Arnold until early 2029, more than three years after Google signed on. The uprade structure Google just signed with Constellation skips that entirely -- the reactors are already licensed, already running, and already connected to PJM. That's the real reason 2028 is realistic here and wasn't for the SMR or restart bets: this deal buys more output from something that already works, not a new thing that has to be proven first.

"Uprade" is a real, decades-old regulatory category, not marketing language -- the Nuclear Regulatory Commission has approved them at US reactors for years through a license amendment rather than a new construction permit. The smallest class, a measurement-uncertainty-recapture uprate, typically adds only 1-to-2% of power by installing more precise flow instrumentation; larger "stretch" and "extended" uprates can add considerably more by upgrading turbines, generators and cooling systems at a plant that keeps running throughout. Neither Google nor Constellation has specified which category of uprate applies at which of the 11 reactors, so the regulatory path -- and how long the NRC actually takes to approve it -- is the detail to watch between now and 2028, not a formality to assume.

The grid this capacity lands on is itself part of the story. PJM, the regional transmission organization coordinating power across those 13 states and Washington, D.C., has warned in its own recent capacity auctions that reserve margins are tightening as data-center demand grows faster than new generation comes online -- a big part of why a hyperscaler paying to extract more output from reactors already connected to that grid reads as a different kind of deal than a company building an isolated power plant to run its own servers off-grid. The five-year Google Cloud/Gemini Enterprise technology alliance layered onto the power agreement is Constellation's attempt to speed that regulatory and engineering path along: Google's tools are pitched at site selection and permitting work, continuous asset-health monitoring on the uprated units, and grid-security infrastructure, rather than anything that touches how the reactors themselves generate power.

The deal also sustains roughly 4,400 existing jobs at the 11 plants and is expected to create about 7,200 construction jobs during the uprade work itself -- a labor footprint that is, again, a function of upgrading existing sites rather than building new ones, which is typically a smaller and shorter-lived construction project than standing up a reactor from scratch.

None of this is Google's first nuclear bet, and it won't be its last -- the company says it has now secured 3,590 MW from Constellation alone, a quarter of it brand new capacity. What makes Tuesday's deal different from the four before it is schedule, not scale: it's the one most likely to actually put new electrons on the grid before the decade is out, because it asked an operating reactor to do more instead of asking a new one to exist.

We're committed to meeting our growth responsibly by actively investing in clean, reliable power that brings new capacity to our nation's grids.