Constellation says Google's 20 year contract enables more than $4.3 billion of uprates, at least $4.8 million per megawatt. The Amazon deal signed Sept. 30 implies at least $15.8 million.
Google's nuclear contract with Constellation buys added capacity at roughly a third of the price per megawatt that Amazon agreed a week earlier. A 20 year power purchase agreement covers 890 megawatts of new capacity at 11 reactors in Illinois, Pennsylvania and New Jersey, which Constellation says enables more than $4.3 billion of investment, or at least $4.8 million a megawatt. On Sept. 30 Constellation announced a 20 year agreement with Amazon adding about 190 megawatts at Calvert Cliffs in Maryland for more than $3 billion, online between 2030 and 2032. That is at least $15.8 million a megawatt.
The gap shows where hyperscaler money goes first. Uprating a running plant extracts more output from turbines, steam generators and digital controls that already exist, which is the work Google describes. New capacity at a site costs several times as much per megawatt. The comparison is rough, because the Amazon contract also supports a 20 year relicensing of the plant, which softens the real ratio. Buyers will still fund the cheapest added megawatts first, and the cheapest are small.
Amazon's deal has the same shape. Constellation's Amazon release pairs 690 megawatts of contracted power with about 190 megawatts of new capacity, 28 percent. Google's ratio is 25 percent. In both, the headline megawatts overstate what is being added by roughly three quarters, and the buyers pay a long term premium for the part that is new.
The ordering matters for a grid that is short today: Google's first uprate is due in 2028 and Amazon's expansion does not arrive before 2030. Of the 3,590 megawatts in Google's headline, 2,700 come from plants already running under a separate 15 year agreement and only 890 are new. Spread across 11 units, that averages about 81 megawatts a unit. The joint announcement plans all of it before the end of 2032, and Constellation expects the first uprate in 2028. Constellation's chief executive, Joe Dominguez, called the Google deal a model for how technology companies and the energy industry can work together, "funded by private entities," according to the same report. The package also includes a five year technology alliance, which adds little to the supply arithmetic but makes the buyer a partner in the program: it has agreed to flex its load and share technology work with the seller.
The agreements cover 2,700 MW from existing reactors and 890 MW from planned uprates. Only the uprate tranche adds plant capacity.
View data
| Contracted capacity | Capacity |
|---|---|
| Existing reactor output | 2700 |
| Planned uprates | 890 |
Utility Dive reported that PJM is short about 6.8 gigawatts of new capacity and that the grid operator delayed a planned reliability backstop auction after FERC only partly approved it on Sept. 29. Against a gap of 6,800 megawatts, Google's 890 is 13 percent. Add Amazon's 190 and the two deals cover 16 percent, leaving about 5.7 gigawatts. Constellation presents the Google package as a response to PJM's Bring Your Own Power proposal and says it supports growth without passing costs to residential customers. That is the company's account of who pays, and a rate case is where it would be tested.
Once the shortfall is clear, the structure follows. The longer term, 20 years against 15, attaches to the tranche that needs the money, and the shorter term covers output PJM already counts. Moving existing megawatts from one customer to Google adds nothing to a short grid. Only the uprates do, so Google carries the capital program and also accepts load shaping that lets noncritical consumption be reduced during high stress grid events, paying partly in flexibility.
In 2024 it agreed to buy power from Kairos Power's small modular reactors, with the first intended online by 2030 and more through 2035, up to 500 megawatts, Google said. That was riskier technology. Constellation's block is about 1.8 times that ceiling and begins two years sooner, since an uprate extends an operating plant while a small modular reactor must prove a design, a site and a supply chain at once.
Staging helps: units can come online one at a time between 2028 and 2032, so the program starts relieving the grid before it finishes. Delivery is the risk. Eleven units at six sites is a program, and the schedule depends on engineering work at each plant and a PJM interconnection neither company controls. A long contract supports the investment decision but cannot make a steam generator arrive on time. Uprates are still the cheapest path to new megawatts, so a program that slips can beat the alternatives on price.
The dollar figures measure the investment Constellation says each contract enables, which is the right yardstick for how much new capacity a buyer's commitment brings forward. Neither Google nor Amazon has published its contract price.
The deals read as a price list for other buyers. Existing output is quick and adds nothing. Uprates add supply at about $4.8 million a megawatt, in increments of roughly 81 megawatts, on a schedule that runs to 2032. Expansion at a site, as Amazon is buying, adds more at three times the cost and later. A hyperscaler that wants to add firm power to PJM will find the cheapest increments are the small ones.
By the end of 2028 only one or two of Google's 11 units, and so well under a quarter of the 890 megawatts, should be in service. Even if every megawatt in both deals lands on schedule by 2032, PJM remains short by roughly 5.7 gigawatts, and with the backstop auction delayed, the next hyperscaler contracts will have to look more like Amazon's price per megawatt than Google's. Constellation's own target of a first uprate in 2028 is the date to check this against.