Nuclear Uprates: How Big Tech Is Powering AI with Existing Fleets

Zach Martin

Welcome to NUCLEAR CAPITAL, on KMKT I R Hub Radio. Real assets. Real ideas. I'm Zach Martin, and I'm here with Emily Martin. Last time we laid out the pattern from the first week of October: Google, Amazon, the Vistra loan, all stacking up. Today we go under the hood, into the contracts, the hardware, and the calendar. One note before we start. Every company we mention today, including Constellation and Vistra, is mentioned for informational purposes only. Nothing here is investment advice, and nothing here is a recommendation to buy or sell anything.

Emily Martin

Thanks, Zach. I'll plant my flag early. The smartest AI power story right now isn't a new reactor design. It's the reactors that already exist. They have a fence, a substation, a cooling system, a license, and a workforce. If a hyperscaler wants firm, clean megawatts this decade, the existing fleet is the shortest line between a check and a connected electron.

Zach Martin

I agree with the direction and I'll be the cautious voice on the pace. Let's put the deals on the table. Google signed a twenty year power purchase agreement with Constellation, announced October sixth, supporting eight hundred ninety megawatts of new nuclear capacity across eleven reactors in the PJM region. Constellation says it will invest more than four point three billion dollars in private capital. The first uprate is expected by 2028, and the brief we're working from has the full eight hundred ninety online by the end of 2032. That's the headline. The footnote is that most of those megawatts are four to six years out.

Emily Martin

Four to six years is fast in this business, Zach. Compare it to the alternatives. Our research puts uprate capital at roughly three thousand to five thousand dollars per kilowatt, versus ten thousand to fifteen thousand or more for new builds. Do the back of the envelope on Google's deal: four point three billion over eight hundred ninety megawatts is around forty eight hundred dollars a kilowatt. Right in that band. And one caveat, since you like caveats: Constellation says that spend also covers modernizing turbines, steam generators, and digital controls, so it isn't a pure per megawatt price tag.

Zach Martin

I do like caveats, and thank you for saying it before I did. Here's another one. Constellation hasn't formally announced which eleven units are in the deal. A company spokesman told Nuclear News the plants include Braidwood, Byron, LaSalle, and Quad Cities in Illinois, Limerick in Pennsylvania, and Salem in New Jersey, where Constellation owns forty three percent. That's six sites for eleven units. And the Nuclear Regulatory Commission's list of expected uprate applications shows Constellation filings over the next few years, one of them in the fourth quarter of 2028, with no plant names attached. So I'd like to see how a first uprate delivered by 2028 squares with that list.

Emily Martin

That's a fair question, but I wouldn't treat it as a contradiction. The deal has several moving parts, and I'm not going to invent an explanation the sources don't give us. What I will say is that this isn't a one off. Meta's January deal with Vistra covers more than twenty six hundred megawatts, and four hundred thirty three of those are incremental uprate capacity at Perry, Davis-Besse, and Beaver Valley. Amazon signed for about six hundred ninety megawatts at Calvert Cliffs, including a one hundred ninety megawatt uprate, delivered between 2030 and 2032. Three hyperscalers, one playbook. Pay for the upgrade, lock in the output for twenty years.

Zach Martin

Let's talk about that Amazon name, because it's the through line to the structure question. Amazon's earlier try, buying power directly from Talen's Susquehanna plant behind the meter, got rejected by FERC in November 2024, and the commission doubled down on that rejection in April 2025. Utilities including AEP and Exelon argued it could shift up to a hundred forty million dollars a year in transmission costs to PJM ratepayers. The new deals go the other way. The power goes onto the grid, front of the meter. Vistra's own release says the electricity continues to go to the grid for all electricity users.

Emily Martin

Which I think is the grown up answer. Behind the meter was a legal fight. Front of the meter is a contract. And look at the framing in the Google announcement: both companies say it aligns with the Ratepayer Protection Pledge and the Department of Energy's UPRISE program, with the capital private and not passed through to household bills. That's a political design as much as a financial one, especially with data centers becoming a midterm talking point.

Zach Martin

Here's where I'll push back, though. Front of the meter means Google isn't getting dedicated electrons. It's getting a contractual claim on output that flows to everyone. For the hyperscaler that's a cleaner path, but it's also a more abstract product. And the colocation argument isn't dead on the merits. The Nuclear Energy Institute paper by former PJM executive Michael Kormos argued the cost shift claim is misplaced. His point was that a colocated data center doesn't take grid service, so there are no costs to shift. FERC didn't buy it in that case. But it tells me the debate is paused, not settled.

Emily Martin

Paused, I'll give you. But notice what the market did while it was paused. Capital moved to the structure that clears. If you're a CFO deploying four billion dollars, you take the path with the fewest regulatory unknowns. That's not abstract, that's discipline. It's actually your kind of lesson, Zach. Small, compounding, boring steps beat the heroic leap.

Zach Martin

Ha, I'll take that, and then use it against you. Boring and compounding also means the megawatts arrive slowly. The Vistra loan is a good example. The Department of Energy's Office of Energy Dominance Financing, formerly the Loan Programs Office, announced a conditional commitment of up to four point two billion dollars. Conditional. Vistra has to satisfy technical, legal, environmental, and financial conditions before there's a definitive agreement and any money moves. That supports four hundred thirty three megawatts of uprates and preserves about four gigawatts of existing baseload.

Emily Martin

And that's the number I'd underline: four gigawatts preserved. Everyone fixates on the four hundred thirty three. But the Vistra plants were, in the words of their own chief strategy officer, on a path to retirement as recently as 2020. The offtake contract and the loan let Vistra plan subsequent license renewals, twenty more years each. So the loan isn't an apples to apples price for four hundred thirty three megawatts. It's financing for expanding and keeping four gigawatts running.

Zach Martin

Okay, that's a concession I'll make, and it changes my framing. The value in these deals isn't only the new megawatts. It's the retirement that doesn't happen. I'd still say the market shouldn't confuse the two. Preserved capacity keeps the reserve margin from falling. New capacity actually raises it. Those are different jobs.