Announced vs. Signed: How Much AI Nuclear Power Is Real?

Zach Martin

Welcome to NUCLEAR CAPITAL, a KMKT I R Hub Radio exclusive. I'm Zach Martin.

Emily Martin

And I'm Emily Martin. Quick note before we start. Every company we discuss today, including Constellation, Vistra, Meta, Google, Amazon, AEP, Oklo, TerraPower and X-energy, is mentioned for informational analysis only. Nothing here is investment advice or a recommendation to buy or sell any security.

Zach Martin

Good. And today's topic is a gut check. We've all seen the headlines about gigawatts and gigawatts of nuclear power for AI. So here's the question I want to chase for the next twenty minutes. How much of it is real? And I mean real the way a banker means it, with a signature and a binding obligation.

Emily Martin

Here's the number that frames everything. As of mid October 2026 reporting, public tracking of nuclear arrangements with AI and hyperscale data center buyers shows roughly 40.4 gigawatts in announced agreements. Those are memorandums of understanding, letters of intent, joint development agreements and master frameworks. Against that, about 10.8 gigawatts sits under signed, binding contracts. Power purchase agreements, offtake commitments, that kind of paper.

Zach Martin

So for every one megawatt under a binding contract, about 3.8 megawatts are announced but nonbinding. And if you do the math on the whole stack, signed capacity is a bit over a fifth of the headline total. Under 22 percent.

Emily Martin

Right. And I'll say up front that I don't read that as a scandal. I read it as a ladder. Some rungs are signed, some are being built.

Zach Martin

I'll push on that, because I'm the cautious one here. A ladder implies you're climbing. A lot of these announcements might just be a picture of a ladder. If project financiers and bondholders can't underwrite a letter of intent, then roughly 78 percent of the headline number can't raise project debt today. That's not a rung. That's a wish list.

Emily Martin

Except it's not a wish list, Zach, and here's why. Nonbinding paper does a job. It sends a demand signal to the supply chain vendors and to regulators. It tells a fuel supplier or a turbine manufacturer that somebody serious wants this. That's not deploying capital, I'll grant you, but it's what makes the later binding contract possible.

Zach Martin

Fair, but a signal isn't a payment. When I think about discipline, and I think about it a lot, the question is always what's committed versus what's hoped for. A hope can be sincere and still not be bankable.

Emily Martin

I'll concede that point fully. Binding PPAs with creditworthy hyperscale buyers are what secure debt and lower the weighted average cost of capital. A framework doesn't do that. So the 3.8 to 1 tells you where the capital can actually flow right now. I just don't want listeners to hear it as the other 40 gigawatts being fake.

Zach Martin

Agreed. Not fake, unfunded. Let's make it concrete, because the ratio is very uneven depending on who the buyer is. Walk me through it.

Emily Martin

Start with Microsoft. In this tracking it's roughly 100 percent signed. It's anchored by the binding 20 year PPA from September 2024 with Constellation for 835 megawatts from the restarted Three Mile Island Unit 1, now renamed the Crane Clean Energy Center. Target commercial operation is 2028, and the NRC issued an environmental assessment finding in September 2026.

Zach Martin

One buyer, one plant, one contract. Simple. Then Amazon, about 86 percent signed, and the newest piece is the September 30, 2026 deal with Constellation at Calvert Cliffs in Maryland.

Emily Martin

It's a 20 year PPA for 690 megawatts from the plant in Lusby, Maryland. The package funds a 190 megawatt uprate, backed by more than 3 billion dollars in capital investment, alongside relicensing and life extension commitments. So you've got existing output locked up and new megawatts from the same site.

Zach Martin

Then it gets less tidy. Google is about 56 percent signed. The anchor is the October 2026 deal with Constellation. Let's spend a minute there, because it's the biggest recent example of a binding deal.

Emily Martin

Sure. It's a 20 year PPA that funds 890 megawatts of incremental capacity from equipment uprates at 11 existing reactors in Illinois, Pennsylvania and New Jersey. The press release says it requires more than 4.3 billion dollars of investment by Constellation. First capacity is expected by 2028, and the brief we're working from has full deployment targeted before year end 2032.

Zach Martin

And separately there's a 15 year agreement covering 2,700 megawatts from Constellation's existing PJM fleet. That one matters for a different reason. It isn't new megawatts. It's long term revenue support for keeping those plants running.

Emily Martin

That's the part I find most interesting. The 2,700 is a floor under the existing fleet, and the uprates sit on top of it. Joe Dominguez of Constellation framed it as funded by private entities, and said it can serve as a model for how tech companies and the energy industry work together. That's his framing, and it's a company view, but the structure is clear. Binding paper on both layers.

Zach Martin

Also a five year technology agreement, with Constellation bringing in Google Cloud and Gemini Enterprise tools for things like outage management and power flow modeling. Nice, but that's not what moves the ratio.

Emily Martin

No. What moves Google's number is the unsigned side. A 500 megawatt master development agreement with Kairos Power for Hermes small modular reactors, targeting 2030 to 2035, and a restart framework with NextEra for Duane Arnold in Iowa. Real intent, but not yet the same kind of contract.

Zach Martin

Now Meta, which is the lowest at roughly 48 percent signed, and the most complicated. In January 2026, Vistra disclosed in an SEC filing 20 year PPAs with Meta covering 2,609 megawatts across three plants.

Emily Martin

Break that down. The operating slice is 1,268 megawatts from Perry and 908 from Davis-Besse, both in Ohio, with deliveries ramping from late 2026 to full by the end of 2027. The uprate slice is 433 megawatts. That's 213 at Perry, 80 at Davis-Besse, and 140 at Beaver Valley in Pennsylvania, with deliveries between 2031 and 2034.

Zach Martin

And now I need to be careful about one thing, because there's a lot of loose talk. Early October reporting from Bloomberg and Reuters says the Department of Energy is preparing a loan of roughly 4 billion to 4.2 billion dollars for Vistra, to uprate and modernize those Ohio and Pennsylvania plants. But the only official document in the public record is a DOE Section 106 letter dated September 15, 2026.

Emily Martin

And that letter says DOE is evaluating whether to provide a Title XVII loan guarantee, and only for the Perry uprate project. Construction was anticipated to begin in October 2026, with work like replacing rotors and moisture separator reheaters, an in vessel steam dryer, and an auxiliary cooling tower. So reported is not official. A direct loan and a guarantee are different instruments, and the reporting blends them.

Zach Martin

Which is a mini version of today's whole theme. Reported versus documented. Announced versus signed.

Emily Martin

It is. And the rest of Meta's number is the development stage. A 1.2 gigawatt power campus with Oklo in Pike County, Ohio, and development funding for two 345 megawatt Natrium reactors from TerraPower, so 690 megawatts, with options for up to six more units, another 2.1 gigawatts.

Zach Martin

Add it up and Meta's broader framework reaches up to 6.6 gigawatts. But roughly half of that is signed and half is development stage. That's the 3.8 to 1 ratio in a single company.