Constellation Energy published its second-quarter report on 5 August 2026. We are writing about it today, 6 August.
What was reported
| Quarter | Year ago | Change | |
|---|---|---|---|
| GAAP net income | 513 | 839 | -39% |
| GAAP EPS | $1.42 | $2.67 | -47% |
| Adjusted operating earnings | 920 | 599 | +54% |
| Adjusted EPS | $2.55 | $1.91 | +33% |
In millions of dollars except per-share data
Two numbers on the same quarter, moving in opposite directions.
The explanation
Why the two lines tell different stories
At a power company, the gap between GAAP and adjusted is almost always the same thing: remeasurement of hedging contracts.
A generator sells much of its output forward, under futures contracts. The accounting rule requires those contracts to be remeasured each quarter at current market prices - so when the power price moves, GAAP profit moves with it, even if nothing was bought or sold.
Adjusted operating earnings strip exactly that out, and describe how much the company earned from actually running its plants.
And so, at a power company, the number describing the business is the adjusted one - and it rose 33%.
And the guidance
The company raised full-year adjusted operating earnings guidance to $11.50 to $12.50 per share.
A mid-year guidance raise, at an infrastructure company whose output is known in advance, is not a trivial thing. It says the company is selling its power at a better price than it assumed.
And why this connects to the whole week
Constellation operates the largest nuclear fleet in the United States.
And that is precisely the kind of power an AI data center needs: continuous, stable supply that does not depend on sun or wind. What the industry calls baseload.
And it joins a week in which Caterpillar attributed its power generation growth to data centers, TeraWulf signed a $19 billion contract, and Iron Mountain leased 75 megawatts in a single month.
Whoever sells the power sits at one end of that chain. Whoever buys it sits at the other.
My Angle
A personal opinion of Ilan Abramov - not advice, not a recommendation
This is a report whose headline misleads to the downside, and that is rare.
Most of the reports I read this week looked good in the headline and less good on breakdown. Here it is reversed: GAAP EPS falling 47% looks like a poor quarter, while the figure describing operations rose 33% and guidance was raised.
And this is a known characteristic of power companies. Hedge remeasurement makes the GAAP line so noisy as to be nearly useless. Looking at a power company through GAAP profit is simply looking in the wrong place.
And what I take into the wider thesis: Constellation is one of the most direct ways to gain exposure to the AI build-out without being exposed to which model wins. A data center consumes baseload power whether it runs OpenAI or Anthropic. And nuclear is one of the few sources that can supply it continuously.
What I will check next: how much of the output is already committed under long contracts to data centers. That is the figure separating a power company from an AI infrastructure company.
