Amazon locks in 20 years of nuclear power from Maryland's only nuclear plant

Amazon has signed a 20-year power purchase agreement with Constellation Energy covering 690 megawatts from the Calvert Cliffs nuclear plant in Maryland, the companies said on September 30. The deal gives Constellation the revenue certainty to invest more than $3 billion in plant upgrades, add about 190 megawatts of new output, and relicense two reactors that would otherwise shut down in 2034 and 2036 (Constellation press release).

The mechanism is not "Amazon buys a nuclear plant." Amazon is buying power. The two pressurized-water reactor units at Calvert Cliffs were connected to the grid in 1975 and 1976, and their original 60-year operating licenses expire in the mid-2030s. Without a long-term off-taker willing to underwrite the relicensing and refurbishment costs, Constellation had little reason to file the paperwork. The Amazon commitment changes that math. The plant keeps running, the new capacity gets built, and a 1970s asset gets a second life on a 20-year contract (World Nuclear News).

Calvert Cliffs sits on the Chesapeake Bay and is Maryland's only operating nuclear plant. It supplies roughly 40 percent of the state's total power generation and around 80 percent of its clean power, enough to run the equivalent of about 1.3 million homes. The site employs more than 800 people and contributes about $21 million a year in state and local taxes, which makes its scheduled 2030s shutdown a fiscal and grid problem, not just an environmental one.

The deal also covers a related retail supply agreement to support Amazon operations across the 13-state PJM regional grid, which covers the mid-Atlantic and parts of the Midwest. The contracted output feeds directly into the data-center demand curve that grid operators are being asked to absorb. Keeping an existing 1,790-megawatt nuclear plant running and adding output at the same site is one of the few ways to add firm, low-carbon megawatts without waiting a decade for transmission upgrades and a greenfield reactor.

Constellation first outlined the uprate and 20-year life-extension plan in November 2025. The Amazon PPA is the contracted demand that unlocks the $3 billion check. About 190 megawatts of new emissions-free capacity is scheduled to come online between 2030 and 2032, alongside "state-of-the-art improvements" across the rest of the plant. Constellation CEO Joe Dominguez and AWS Vice President Kerry Person are quoted in the corporate release; the deal value, term, and capacity are primary-source confirmed (Constellation press release).

The deal fits a broader pattern. Amazon was part of a group of major energy users that signed a pledge last year supporting the goal of at least tripling global nuclear capacity by 2050. A 20-year Calvert Cliffs commitment is exactly the kind of long-term, fixed-price contract that pledge implies: a major power user underwriting the firm, low-carbon supply an aging nuclear fleet needs to stay in the money.

The unresolved question is whether this is enough. A 190-megawatt uprate and a 20-year license extension on two 50-year-old reactors is a real addition to the grid, but it is not greenfield build, and it is not a substitute for new baseload capacity where data-center demand is rising fast. The deal also has a conditionality that should not be lost. The 20-year extension is a forward-looking commitment enabled by revenue certainty; the actual relicensing still requires Nuclear Regulatory Commission approval. If the application stalls, the cash does not convert to capacity.

A second question is what additional capacity, if any, will sit at the Calvert site beyond the uprate. The company's release describes "state-of-the-art improvements" across the 1,790-megawatt plant and 190 megawatts of new output, but stops short of naming any greenfield build on the property. Constellation's next earnings call or NRC filing should clarify whether more is planned.

For now, the artifact is a 1970s plant kept running through roughly the 2050s by a 20-year contract signed in 2026. That is a useful data point for the data-center-and-grid story, and a falsifiable one. The test for the next round of hyperscaler nuclear deals is simple: do they fund new build, or do they fund more uprates?