For about fifteen years, the people who plan America's electricity supply had an unusual luxury: the number they planned around barely moved. In 2007, the country's power plants generated a little over 4,000 terawatt-hours of electricity. Then — through a financial crisis, a fracking boom, the entire rise of the smartphone, and a pandemic — output essentially flatlined. New data centers were offset by departing factories and the slow disappearance of the incandescent bulb; efficiency canceled out growth. As late as 2021, the country still generated less power than it had in 2007. You could build a thirty-year plan around a number that did not change, and most utilities did.

That era ended in 2022, and it ended decisively. Generation broke above its old ceiling and the line has bent upward since — a record in 2024, and a higher one in 2025. The country is now producing more electricity than at any point in its history, and the climb is steepening.

The cause is the least surprising customer imaginable: artificial intelligence. Underneath the software, a modern AI data center is an enormous and very particular electricity buyer. It wants power that is large, constant, carbon-free, and locked in for twenty years — because the people writing checks for a multibillion-dollar computing campus need to know the lights will stay on for the life of the loan. A country that spent fifteen years not needing to build much new power is being asked to build a great deal of it, quickly.

This week, that pressure showed up across our universe all at once — not in a single name, but along the entire chain that turns fuel into electricity and electricity into computing.

When a whole chain re-rates at once, the question is not whether to own it but where on the chain to stand. Each spot carries a different risk and a different mechanism. Four of them stood out this week.

Start with the safest place to stand: the equipment. GE Vernova makes the gas turbines and the wind turbines that generate the power, the transformers and switchgear that move it, and through a small-reactor arm it has a foot in the nuclear path too. It does not have to be right about which kind of generation wins the next decade — gas, wind, or nuclear — because it sells the hardware for all of them. That is the toll-booth position: paid on the traffic, indifferent to the route. Our model agrees. Of everything it tracks, GE Vernova has the most attractive setup right now — its growth, its timing, and its price line up at 86 out of 100 on what we call the ActionScore, a single read on how well a name's case fits this moment. It is also, by a wide margin, the model's highest-conviction name. The one caution is the obvious one: no stock in our universe has more momentum behind its price this week, which is another way of saying it is well up its price cycle and you are paying a full price to stand here.

Now step one link down the chain, to the power itself — and to a very different setup. Constellation runs the largest fleet of nuclear reactors in the country. Its electricity is the rare product that is already large, already constant, already carbon-free, and already there, which is why it is the company actually signing twenty-year agreements to sell power to AI operators. Here is the contrast worth noticing. Where GE Vernova's price has raced ahead of its business, Constellation's business has raced ahead of its price. By our model its underlying case is solid — a 74 on that same setup scale, its momentum running well ahead of most of the universe — yet its price cycle still reads cool: early, not yet chased. (Our automated model portfolio opened a Constellation position on June 10.) The same conviction in the data, expressed two completely different ways — one where the crowd has already arrived, one where it has not.

The third place to stand is the least glamorous and the hardest to skip: the wires. A reactor or a turbine is worthless until its power reaches the building, and that last stretch is its own industry — the transformers, the switchgear, the substations, and the crews who build them. These are mostly steady growers rather than dramatic ones: nVent in electrical protection, which happens to be the single largest grid position in our model portfolio; Hubbell in electrical equipment; Quanta and Dycom in the construction itself. None is a wager on a breakthrough. They are a wager that everything above them on the chain has to physically connect to something — and that the connecting has barely started.

The fourth place is the one to size smallest and watch hardest: the reactors that mostly do not exist yet. The small-modular names — NuScale, Nano Nuclear, Oklo — are bets on a design moving from regulatory approval to poured concrete. This is where our model earns its keep precisely because it refuses to flatter them. With little or no revenue to score, their fundamentals read close to zero: NuScale rates a 0 for growth and a 2 out of 100 on the question of whether it is a sound business at a fair price, even as its share price accelerates. The system will not pretend there is a company where there is, so far, a blueprint. That is the definition of a higher-risk, higher-upside position — real optionality on the part of the story that has not happened yet, owned with eyes open and in small size.

The thread tying the four together is a single conditional. IF the AI buildout keeps signing twenty-year power agreements at the pace it set this year, AND WHEN the first small-modular designs move from approval to first concrete, THEN the scarce thing stops being the reactors and becomes the capacity to equip and connect them — and the toll-booth names like GE Vernova are paid first, longest, and regardless of which generation source ultimately wins.

The shift was not subtle, and it was not only about what is rising. The clearest signal this week was a rotation: heat draining out of last year's winners and into the chain that powers this year's.

The week in numbers

What's moving

From the machinery Our automated model portfolio reads the same shift. Its highest-conviction name, by a wide margin, is GE Vernova — it wins almost every head-to-head comparison the ranking engine puts it through. The grid sits among the portfolio's three largest commitments of actual capital, alongside AI hardware and defense. And the freshest move it made in this corner of the market was opening a Constellation position on June 10 — real money behind the operator with power to sell, not only the equipment maker with the highest score.

Scorecard — Constellation (CEG)

Each number is a 0–100 rank against the whole universe; the Heat column is its price-cycle read. The picture: a sound business the market has not yet chased — its appeal and momentum well above average, its growth comfortably so, while the price cycle still reads cool. The setup most people would call "early."

A flat-demand era ending is not a quarter's worth of news; it is a turn in the baseline. Power agreements run twenty years. Reactors take a decade to build and run for sixty. The grid that carries it all is being asked to grow for the first time in a generation, and you cannot will transformers and trained crews into existence on a data center's timeline. The names that sit on that chain were not worth a second look when the country's output sat still. It is not sitting still now.

Flat was the exception, not the rule.

Research and educational analysis only — not financial advice. Every investment carries risk.