Google signs 20-year deal to add 890MW to PJM by uprating 11 existing reactors, shifting from power buyer to capacity provider on a grid 6.8GW short

Google and Constellation Energy have announced an agreement to add 890MW of new nuclear capacity to the PJM grid (Constellation press release, primary source). No new reactors will be built. Instead, the output of 11 reactors already in operation will be raised through power uprates, and Google is backing the cost with a 20-year contract.

What makes the deal significant is its structure more than its size. Until now, most hyperscaler power deals have involved buying output in bulk from existing plants or purchasing renewable energy certificates. This time, Google has committed for 20 years to fund capacity that does not yet exist on the grid. Google, long a buyer of electricity, is starting to become a provider of grid capacity.

Uprates: adding capacity without building new plants

New nuclear plants take a decade or more to go from licensing to operation. Small modular reactors (SMRs) still have little commercial operating history. An uprate, by contrast, raises the output of a reactor that is already licensed and running by replacing or upgrading equipment such as turbines and steam generators. Because the site and the transmission connection already exist, capacity can be added faster and with more certainty than building from scratch.

Still, 890MW across 11 reactors works out to an average of only about 80MW of added output per unit. In effect, the capacity of one large reactor is being spread in small increments across 11 existing units. A breakdown of how many megawatts will be added at which plants and when, and the amount Google will pay, could not be confirmed at the time of writing. This is because the sources could not be fully reviewed, and reported contract values have in the past differed widely from the actual figures. This article therefore limits itself to the three published numbers: 890MW, 11 reactors and 20 years.

Even so, a customer attaching a 20-year commitment to projects that are each small in scale carries real weight. Uprates have been an investment whose returns utilities found hard to predict, because no one knows what power will cost decades from now. A 20-year buyer removes that uncertainty. Google's money helps tip the decision to start work.

A 6.8GW shortfall at PJM: the shortage is capacity, not electricity

PJM is the largest grid operator in the United States, spanning 13 states and the District of Columbia from the East Coast to the Midwest. It includes the data center cluster in Northern Virginia, making it the grid where growth in AI-driven demand shows up most directly. PJM is said to be short about 6.8GW of capacity.

Capacity here means generating capability that is committed to deliver power even during hours of peak demand. It is distinct from the amount of energy produced over a year. Data centers draw a nearly constant load around the clock, so what they need is not certificates from solar plants that generate only during the day, but capacity that can be dispatched at any time. Nuclear uprates fit that requirement well, because they draw from existing facilities the capacity of a source that can run almost continuously, regardless of weather.

On a grid short of capacity, capacity prices rise, and those costs ultimately land on household and business electricity bills. Criticism that data center demand is driving up costs for other customers has been growing across the United States. Adding new capacity to the grid with its own money is also a hyperscaler's answer to that criticism: for the power it uses, it adds capacity to the grid.

Market reaction: reactor owners repriced

According to 247wallst, Constellation shares rose 12% on Oct. 6. Fellow reactor owners Vistra and Talen Energy gained 8% and 7%, respectively (all as reported).

The three companies likely rose together because the market read the deal as something bigger than a single Constellation story. If existing reactors still have room to raise output, and hyperscalers will pay 20 years' worth for that room, then owning reactors becomes more valuable in itself. Beyond the price of power, holding capacity that can be added to the grid is starting to become a product in its own right.

From buyer to provider: hyperscalers change how they secure power

Hyperscalers have so far secured power in two main ways. One is buying finished electricity from utilities or generators under long-term contracts. The other is drawing power directly from a plant next to the data center, which effectively removes that source from the grid. The latter has drawn criticism for taking capacity away from other customers.

This deal differs from both. Rather than competing for power from existing plants, the customer takes on the long-term cost of adding new capacity to the grid. Utility Dive also reported it as new nuclear capacity for PJM. Against the 6.8GW shortfall, 890MW amounts to about 13%, a substantial share for a single contract from a single company.

The AI race is no longer only about model performance and price. It is also a race to secure, quickly and reliably, the power that runs the compute. Waiting for new nuclear plants or SMRs would take too long. So the approach is to draw capacity out of existing reactors and pay for it directly. That is the path Google has chosen. The question now is whether hyperscalers will go beyond being large utility customers and take part in shaping grid capacity. The next things to watch are whether other hyperscalers follow with similar deals and whether the model spreads to grids beyond PJM.