Inside the Google Constellation Nuclear Deal: 3,590 MW and Who Pays for It

Google has contracted for 3,590 megawatts of power from Constellation Energy under a 20-year agreement inside PJM Interconnection, the largest US grid, with new nuclear generation accounting for roughly a quarter of the supply. The Google Constellation nuclear deal pairs a long-dated power purchase agreement with direct corporate funding for upgrades at 11 operating reactors in Illinois, Pennsylvania and New Jersey.

The timing follows the load. PJM serves 67 million people across 13 states and has been absorbing surging demand from data centers that now compete with households and industry for the same capacity. Contracting for two decades makes Google a fixed buyer in that market instead of a spot purchaser exposed to whatever the grid clears.

Only part of the headline number is new generation. The 3,590 MW total covers power supply across the PJM market, while 890 MW of it comes from uprates, the extra output drawn from reactors that already run. That block is scheduled to reach the grid before the end of 2032.

What the Google Constellation Nuclear Deal Covers

The uprate spending is concentrated in one state. About half of Constellation's investment, roughly $2 billion, targets four Illinois plants: Byron, Braidwood, LaSalle and Quad Cities. The remaining work is spread across sites in Pennsylvania and New Jersey, with total physical and digital upgrades valued at about $4.3 billion.

Employment runs alongside the capital. Constellation expects the projects to support approximately 4,400 existing jobs and to create 7,200 new construction roles, spanning craft labor, engineering and outage crews through the upgrade schedule.

Uprates are a narrower engineering exercise than building new reactors. Output rises through modifications to turbine, generator and instrumentation systems instead of new containment structures, which shortens timelines and lowers capital intensity. The ceiling is set by the original plant design: 890 MW spread across 11 units averages roughly 80 MW per reactor, and regulators must clear the changes before the additional output counts.

That arithmetic explains why the deal leans on sites that already exist. Uprates deliver firm capacity years sooner than a new reactor, which in the United States would take the better part of a decade to license and build.

Measured against the full 3,590 MW, the new nuclear component is a minority of the package. The rest is existing carbon-free output from Constellation's fleet, so the agreement is largely about securing firm supply rather than commissioning new generation. That distinction matters for anyone tracking whether AI demand adds capacity to the grid or reallocates what already exists.

For Constellation, the practical value is a revenue floor. A 20-year contract with an investment-grade counterparty gives the company a basis for committing $4.3 billion without a rate case, a cost recovery mechanism, or merchant assumptions about where power prices sit in the 2030s.

The commitment itself is valued above $1 billion, and it bundles existing generation with the uprate block rather than resting on new construction alone. For a buyer with round-the-clock load, that mix is the point. Nuclear output is available at any hour, which sidesteps the storage and curtailment questions that follow intermittent supply.

From Power Buyer to Power Financier

The structure of the arrangement is the part that travels. Costs for new capacity land on one large corporate buyer rather than being distributed across utility ratepayers, separating the entity paying for generation from the customers consuming it. Google's earlier Georgia agreement followed the same pattern, enabling $900 million in customer benefits at no cost to other ratepayers.

Google's enabled nuclear capacity in the United States now exceeds 1.5 GW. That total includes the Duane Arnold Energy Center restart in Iowa, which accounts for more than 600 MW, and the new Constellation commitment is the largest single addition to it.

What makes the arrangement unusual is the direction of the money. Instead of signing a conventional offtake and leaving construction risk with the generator, Google is funding the upgrades itself, which puts it close to an equity-style partner in assets it does not own. That position is likely to be tested as more AI capacity comes online and grid operators weigh who should pay for the transmission and capacity needed to serve it.

Constellation's other customers keep drawing from the same fleet. Because uprates add output instead of redirecting it, the new megawatts are incremental to the supply stack rather than a transfer of power other buyers already rely on, which is what keeps the arrangement defensible in a market where capacity is tight.

Investors read the announcement as a re-rating of the uprate opportunity. Constellation shares rose about 3% in premarket trading to roughly $276, a level that implies the market treats contracted demand as insulation from the capital risk that ordinarily attaches to generation projects.

A software layer sits on top of the hardware. Google is deploying Gemini Enterprise to help Constellation optimize operations across its plants, which places the buyer's AI inside the workflows of the operator it depends on.

Digital upgrades are the less visible half of the $4.3 billion program. Monitoring and analytics work of that kind is what makes an enterprise AI deployment useful inside a plant, and it links the software contract to hardware the buyer helped finance. For Google, that is a reference deployment in an industry where vendors face long qualification cycles.

Capacity auctions are the mechanism PJM uses to price strain on the system, and a direct contract lets a buyer step around them. Google gains volume and cost certainty; PJM gains 890 MW of firm, carbon-free capacity it would otherwise have to source elsewhere, at a moment when data center forecasts keep climbing.

The Regulatory Clock

The deal arrives roughly a week before the Federal Energy Regulatory Commission decides whether data centers must bring their own power to the grid. The ruling will determine how far large loads can co-locate behind the meter and sidestep the cost-allocation rules that spread transmission and capacity expenses across ratepayers in the territory.

A permissive outcome would make self-supply easier for hyperscalers and weaken the case for long-term contracts of this kind. A restrictive one pushes the other way, making arrangements in which a single buyer underwrites generation the path of least resistance. Either result keeps pressure on PJM's capacity market, where data center forecasts have been repricing the value of firm, carbon-free output.

The 20-year tenor carries its own logic. Nuclear plants run for decades, and matching contract length to asset life lets an operator amortize upgrades against committed revenue. Data center campuses depreciate faster than reactors, so Google is accepting duration risk that most corporate power buyers avoid.

Two milestones will show whether the model holds. One is the FERC decision expected within days of the announcement. The other is the uprate schedule itself, which has to deliver 890 MW across 11 units in three states by the end of 2032.

Why this matters

Hyperscalers are shifting from buying power to financing it. Google's contract moves the cost of new nuclear capacity onto a single corporate balance sheet and puts its own software inside the plants it relies on. For the 67 million people served by PJM, that changes who carries the risk of meeting rising demand. For Constellation and its peers, it sets a template that other large buyers will be measured against.

Sources

Why we're backing America's existing nuclear plants

Related Articles

• Google's Nuclear Uprate Tariff Puts AI's Power Bill on Georgia's Rates
• Washington's $1.9B bet on the Duane Arnold nuclear restart
• Oracle's Project Jupiter Force Majeure Notice Tests AI Data Center Financing

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.