Why Existing Nuclear Plants Can Beat New Reactors

Zach Martin

Welcome to NUCLEAR CAPITAL on I R Hub Radio. I'm Zach Martin, and it's Monday, October 5, 2026.

Emily Martin

And I'm Emily Martin. Before anything else, a standing note. This is a capital markets briefing for informed listeners. It is informational only. Every company we mention today, including Constellation, Amazon, and Cameco, is mentioned for information only. None of it is a recommendation to buy or sell anything.

Zach Martin

Good. So here's today's question, and I think it's a fun one. Everybody talks about the new reactor race. New designs, new sites, new construction. But the fastest new carbon free power for the AI economy might come from a plant that's already running, and that's already wired into the grid. Emily, why can an uprate at an operating plant show up before a brand new reactor?

Emily Martin

Because the hardest parts of nuclear are the parts an operating plant has already finished. Think about what a greenfield project has to do. It has to win site permits, secure cooling water, build a security footprint, build a switchyard, and get an interconnection to the grid. An operating station already has all of that. It has environmental permits, licensed cooling water infrastructure, a security footprint, a switchyard, and a grid connection that's been carrying power for decades.

Zach Martin

So it's like adding a second story to a house that already has the foundation, the plumbing, and the driveway.

Emily Martin

That's a fair picture, with one caution. A second story is still a serious construction job. We'll get to the timeline, because it's not instant. But you're starting from a very different place than an empty field.

Zach Martin

And the example today is the one that landed right at the end of last month. Around September 30, Amazon and Constellation announced a 20 year power purchase agreement supporting the expansion of Maryland's Calvert Cliffs nuclear plant. Let's lay out what's been reported, carefully.

Emily Martin

As reported in Constellation's announcement, the agreement is a 20 year PPA covering about 690 megawatts. Included in that is approximately 190 megawatts of additional generating capacity, and that addition is targeting operation in 2030 to 2032. The agreement is reported to support more than 3 billion dollars in plant improvements and expansion. And it comes alongside efforts to renew the plant's operating licenses for another 20 years.

Zach Martin

Now, let's be clear about what this deal is not, because the headlines can blur things. It's not a small modular reactor story. Nobody is building a new reactor design here. And it's not the Three Mile Island restart story either, where a shuttered unit is being brought back. This is Big Tech offtake meeting the existing U.S. fleet.

Emily Martin

Right. And the arithmetic is worth doing out loud. If the whole agreement covers about 690 megawatts and about 190 of that is additional capacity, then by simple subtraction roughly 500 megawatts is the plant's existing output being contracted. So most of the deal is a buyer committing to power that already exists, and a smaller slice is new.

Zach Martin

That matters, because the reporting we've been reading makes this distinction itself. Buying existing electricity does not by itself add generation. So the part that changes the physical grid is that 190 megawatt addition, and the license renewal efforts that keep the plant running longer.

Emily Martin

Exactly. And that's where the contract earns its place. Let's talk about why a 20 year offtake changes the financing case. Historically, a merchant nuclear operator sells into wholesale power markets. Prices move around. That makes it hard to commit to a multi billion dollar upgrade program, because you can't be sure what the power will sell for in year eight or year fifteen.

Zach Martin

It's like a farmer deciding whether to buy a new irrigation system when he doesn't know what the crop will sell for. Hard to borrow against. But if a creditworthy buyer says, I'll take what you grow for twenty years, now the math changes.

Emily Martin

That's the mechanism. A 20 year agreement with an investment grade buyer like Amazon turns a variable revenue stream into a more predictable, contracted one. That's the framing in the reporting too, that longer term AI electricity agreements give operators greater revenue certainty for upgrades and longer operation. It's an interpretation of how the economics work, not a disclosed term of the contract. We don't have the pricing, and we shouldn't guess at it.

Zach Martin

Good point. We don't know the price per megawatt hour. What we know is the structure: long term, large buyer, and more than 3 billion dollars of improvements and expansion tied to it.

Emily Martin

And that structure isn't unique to this deal. The International Energy Agency reported in April 2026 that power purchase agreements between existing nuclear generators and AI related companies covered about 7.1 gigawatts from 2024 through the first quarter of 2026. The IEA said most had reached final investment approval, and described them as firm commitments, whether to buy existing output, restart reactors, upgrade plants, or extend operating lives. Buyer commitments of up to 25 years were part of that picture.

Zach Martin

Now, I want to flag how easy it is to misread that 7.1 number. It sounds like 7.1 gigawatts of new reactors.

Emily Martin

It isn't. It's contracted power, not equivalent new capacity. A lot of those deals lock up electricity from plants that were already generating. So think of it as context. It tells you Big Tech is lining up long contracts with the existing fleet. It doesn't tell you how many new megawatts that created. Calvert Cliffs is interesting precisely because a defined piece of it, that 190 megawatts, is described as additional capacity.

Zach Martin

So let's put 190 megawatts next to the thing driving all of this, which is data center load. What does 190 actually mean?

Emily Martin

A common rule of thumb is that a large hyperscale AI campus can run somewhere between 100 and 300 megawatts. That's a general industry range, not a figure from this deal. So 190 megawatts of around the clock power is roughly in the neighborhood of one big campus. That's meaningful for a single plant. But zoom out, and the IEA projects global data center electricity consumption at 950 terawatt hours in 2030, up from 485 in 2025, and it projects consumption by AI focused data centers tripling over that period.

Zach Martin

So 190 megawatts is real, it's useful, it's firm. But it's one campus worth of power against a global load that's roughly doubling in five years.

Emily Martin

Yes. Which is why you'll hear the same pattern repeated across the fleet, if the economics keep working. It's a lot of small to medium increments from existing sites, and each one is quicker than starting from scratch. But none of it is a silver bullet.