Google Bought 3.6 Gigawatts of Power. Only 890 Megawatts of It Is New.

Tuesday's announcement from Google and Constellation Energy led with a number big enough to stop a scrolling thumb: 3,590 megawatts. A 20-year power purchase agreement. Nearly a gigawatt of new nuclear capacity. More than $4.3 billion of investment. A grid serving 67 million people.

That is the version that got the headlines. Here is the version that matters to anyone trying to buy firm power on a schedule. Roughly a quarter of the 3,590 megawatts is new supply. The other 2,700 megawatts come from plants that are already running.

Two agreements, one press release

The structure is worth separating, because the two halves do different jobs.

The first is a 20-year power purchase agreement covering 890 megawatts of new nuclear capacity. Constellation will fund upgrades at 11 nuclear units across six sites in Illinois, Pennsylvania and New Jersey. The first uprated unit is targeted for 2028, and Google says the full 890 megawatts arrives before the end of 2032. Constellation puts the investment at more than $4.3 billion, with roughly 4,400 existing jobs sustained and about 7,200 construction jobs during the build.

The second is a 15-year energy supply agreement for 2,700 megawatts from Constellation's existing PJM fleet. Nothing gets built for this number. Constellation describes it as revenue certainty for assets that are already operating: "Ensuring continued operations through a 2,700 MW, 15-year agreement that ensures existing assets remain economically viable without taking resources off of the grid."

Joe Dominguez, Constellation's chairman and CEO, framed the package as a model for private funding of grid capacity: "This long-term clean energy collaboration with Google can serve as a model for how technology companies and the energy industry can work together to responsibly develop the digital economy and invest in our nation's energy infrastructure in a way that delivers grid-wide benefits for all, funded by private entities."

Amanda Peterson Corio, Google's global head of energy and power, tied it to affordability and jobs: "Our agreement with Constellation to fund nuclear reactor uprates will strengthen the PJM grid, which serves 67 million people, all while protecting energy affordability and supporting local union jobs."

The fastest new megawatt is the one already built

An uprate does not add a reactor. It raises the licensed thermal power of one that already exists, usually by replacing or modernizing equipment on the non-nuclear side of the plant: turbines, steam generators, condensate pumps, transformers, digital control systems. The reactor core runs hotter, so it makes more steam, and the balance of plant has to handle the extra flow.

The Nuclear Regulatory Commission sorts uprates into three tiers. Measurement uncertainty recapture uprates recover less than 2% by improving how precisely reactor power is measured. Stretch uprates, typically up to 7%, adjust instrumentation setpoints and operating procedures without major hardware changes. Extended power uprates can reach 20% and are the capital-intensive ones, often requiring new turbines, generators and transformers. The NRC targets a six-month review for a measurement recapture application, nine months for a stretch uprate and 12 months for an extended one, all handled as amendments to an existing operating license rather than a new one.

The cost comparison explains why hyperscalers keep reaching for this lever. TVA's extended power uprate at Browns Ferry, completed in 2019, cost about $475 million and added 465 megawatts across three units, or roughly $1 million per added megawatt. Idaho National Laboratory places new large light-water reactor construction near $10 million per megawatt as a planning figure. Those numbers come from different years and different scopes, so the comparison is directional rather than exact. The direction is not subtle. An uprate also skips the multi-year queue for a new grid interconnection, which is often the real bottleneck.

Google and Constellation are not alone in this. Constellation signed a 20-year agreement with Amazon on September 30 covering a 190-megawatt expansion at Calvert Cliffs in Maryland, backed by more than $3 billion. Vistra has a conditional Department of Energy loan commitment of up to $4.2 billion behind uprates in Ohio and Pennsylvania. The Department of Energy's UPRISE program is built around exactly this idea.

Where the story gets harder

Three things are worth holding onto before treating this as a solved problem.

The first is that uprates are a margin play, not a buildout. The NRC's expected-application list runs to about 2.4 gigawatts of potential added capacity through 2032. PJM alone is short roughly 6.8 gigawatts of new capacity, according to Utility Dive, and is still working out how to procure it. Extracting more from the existing fleet helps at the edges. It does not close a gap of that size.

The second is execution. Federal backing lowers the cost of capital and signals policy commitment. It does not prevent overruns. Vogtle Units 3 and 4 in Georgia came in above $30 billion against an original budget roughly half that, years late. Any uprate program bundled with life-extension and plant-wide refurbishment can drift the same way. Google's own timeline stretches 890 megawatts across six years.

The third, and the one likely to generate the most friction, is who pays. The 2,700-megawatt agreement keeps existing plants open and paid. That is a cost, and the fight over how it lands on residential bills is already running. In July, ratepayer advocates from Delaware, Illinois, Maryland and Ohio, plus Pennsylvania's consumer advocate, told the Federal Energy Regulatory Commission that its framework for large loads does not stop data center transmission costs from shifting to other customers, as Utility Dive reported. Earthjustice and Energy Futures Group published a report in September cataloguing the same problem across ten cost categories, from fuel price pressure to stranded-asset risk. PJM's own reliability backstop procurement is in limbo after FERC's September 29 order suspended it for five months and opened a section 206 proceeding.

Google is presenting the deal as an answer to that criticism, aligning it with PJM's "Bring Your Own Power" proposal and the White House's ratepayer protection pledge, and describing it as bringing new capacity to the grid without passing costs to residential customers. Whether that accounting holds up across a 20-year term and a 15-year term is a separate question, and not one the press release settles.

What the deal actually says

Strip the megawatt count back and the message is about constraint. Google expects to spend as much as $205 billion on data centers and compute this year, according to The Register. Constellation's shares rose almost 14% on the announcement. Google now says its agreements have enabled more than 1.5 gigawatts of new nuclear capacity in the U.S. through uprates and restarts, with this deal the largest single piece.

That is the shape of the market right now. The compute is not the scarce input. The electrons are. The fastest way to get more of them is to spend money on plants that are already standing rather than waiting a decade for ones that are not.

The contract structure is where the engineering meets the risk. A 20-year offtake used to underwrite a capital upgrade is a supply agreement before it is an energy agreement. It has to price the build, sequence the outages, absorb the cost overruns and survive a regulatory fight that runs for the life of the deal. Google is acting as anchor customer for physical work at 11 reactors, funded on the strength of a long-term purchase commitment. That is a procurement and integration problem wearing a power purchase agreement's clothes, and it is the same class of problem hardware companies hit when demand outruns the capacity to make the thing they need.

DMC works with companies navigating exactly those constraints: sourcing strategy, cost modeling against long lead times, and production ramp planning when the demand curve and the supply curve are moving at different speeds. If you are structuring a supply commitment to underwrite capacity, whether that capacity is a reactor or a fabrication line, let's talk.