Read through the Big Tech nuclear contracts after the CEG/Google deal. Almost none of them just buy existing power

CEG popped again yesterday on the Google deal, a week after the Amazon one. Every take I see is basically "AI needs power, buy anything with a reactor" so I actually went and read the deal announcements.

Neither deal is Big Tech just buying power that already exists. Amazon's 690 MW at Calvert Cliffs comes with \~190 MW of uprates (basically new equipment so the same reactors put out more) and money toward renewing two licences. Google's is 890 MW of uprates plus a big 2,700 MW financial contract where the power stays on the grid.

Then I went back to 2024 and checked every deal between a big tech company and CEG, VST, TLN, NEE or OKLO. I count 9. Seven of them add new capacity somehow, either a restart like Crane, uprates, or Oklo's new reactors for Meta. The two that don't are Talen and Vistra's Comanche Peak, and both are from 2025.

That lines up with the White House pledge from March, where Amazon, Google, Meta, Microsoft and a few others said they'd buy from "new or otherwise additive" plants and pay the full cost so it doesn't hit people's power bills.

So my read is that the reactor count matters less than how much room a company has left to add. CEG has a lot and just used it twice. VST has done it once. Talen is the one that bugs me, since it has one nuclear plant and Amazon already takes about 85% of its share.

Could easily be wrong though. CEG sold \~890 MW of plain existing power to unnamed buyers in Q2, and Google's existing chunk is 3x the new part, so maybe a small uprate is all it takes to get a deal done. And nobody's disclosed prices, so the $7B+ CEG is spending might just earn a boring utility return.

Does anyone know who those Q2 buyers were?

I put the full list of deals with sources on my site (InvestMoat) if anyone wants it: https://investmoat.com/research/constellation-and-the-megawatt-a-hyperscaler-has-to-add

fwiw at InvestMoat we only hold CEG out of these five. It has the biggest fleet nobody can copy, it's already locked up with several big buyers, and it still has room to grow. VST is close but not quite there, NEE grows too slowly, TLN leans too much on one plant and one buyer, and OKLO doesn't have a running reactor yet.