Amazon just signed a 20-year deal for Maryland's only nuclear plant

Constellation and Amazon announced on Wednesday a 20-year power purchase agreement built around the Calvert Cliffs Clean Energy Center in Lusby, Maryland — the state's only nuclear plant and its largest source of clean energy.

The agreement covers 690 megawatts of the plant's 1,790-megawatt output. It also enables what Constellation describes as more than $3 billion of Maryland infrastructure investment: improvements across the whole facility plus an uprate adding approximately 190 megawatts of new emissions-free generating capacity, phased in between 2030 and 2032. After the upgrade the plant's capacity reaches roughly 1,980 megawatts.

Alongside the PPA, the companies signed a retail supply agreement supporting Amazon's operations throughout the 13-state PJM market, which Constellation says will also help Amazon manage regional energy costs.

Why the structure matters more than the megawatts

One hundred ninety megawatts is a small number in a grid conversation. Maryland Matters puts it at roughly enough to power about 147,000 American homes. Against the scale of current data-center demand forecasts, it is a rounding error.

The mechanism is the story. Amazon is not buying a share of an existing plant's output and calling it clean power. Its long-term commitment is what makes the uprate financeable, and what Constellation says gives it the revenue certainty to pursue relicensing for another 20 years. The NRC lists the two reactors' current licenses as expiring in 2034 and 2036. Constellation also frames the commitment as support for developing new clean generation at the same site.

That is a different transaction from the co-location deals that drew regulatory objections over the past two years, where a hyperscaler sought to take power directly behind the meter and critics argued the cost of grid service shifted to everyone else. Here, Constellation is explicit that all of Calvert Cliffs' electricity continues flowing into the PJM regional grid exactly as it does now.

The constraint this is designed to answer

Through 2026 the binding limit on AI infrastructure has moved steadily up the stack: first chips, then advanced packaging, then grid interconnection and firm power. Announcing a data center is easy. Securing two decades of around-the-clock carbon-free electricity in a load zone that already has queue congestion is not.

Nuclear is attractive in that context for an unglamorous reason — capacity factor. A reactor runs near continuously, which matches a data center's flat load curve in a way intermittent generation cannot without storage. The supply problem is that almost no new U.S. nuclear capacity can be delivered this decade. Uprates at existing plants, and keeping existing plants licensed, are the only levers that work on a 2030-2032 timeline.

Which is precisely what this deal pulls: an uprate plus a relicensing path, paid for by a counterparty that needs the power.

For Maryland

The state keeps its largest clean-energy asset running past the mid-2030s and gets $3 billion of capital spending at a single site, with the construction employment that implies. Financial specifics remain confidential, so the ratepayer question — whether a long-term contract at this scale affects what everyone else in PJM pays — cannot be settled from the disclosures.

That is worth stating plainly rather than assuming away. PJM capacity prices have been a live political issue in several states, and a 690-megawatt bilateral contract is large enough to matter to the margin even when the electrons keep flowing to the same grid.

What it signals for the sector

CEG was up about 2.9% in Thursday's pre-market. The more durable signal is in the contract design, which several other operators of existing nuclear fleets can copy: sell a block of existing output on a 20-year term, use the certainty to fund an uprate and a licence renewal, keep the output in the wholesale market, and let the hyperscaler's balance sheet carry the duration risk.

Watch three things from here. Whether regulators treat grid-connected long-term PPAs more favourably than behind-the-meter co-location. Whether other operators of existing plants announce comparable uprate-plus-relicensing packages. And whether these commitments keep getting signed now that long-term borrowing costs sit at 24-year highs, which raises the hurdle rate on every multi-decade energy project being modelled this quarter.

Invested Alpha is a publisher, not an adviser; nothing here is a recommendation. The point is narrower: power procurement has become a disclosed, datable part of the AI capex story, and this is the template being used.