Something strange happened in April.

A nuclear company, priceing its shares 21 percent above the top of there initial range, raised $1.02 billion on its first day of trading, 15 times (!) oversubscribed, surging 31 percent out of the gate. The same company had try to go public via SPAC in 2023 and the whole thing fell apart, the deal terminated, the partner firm liquidated, the market said no.

What changed is that the worlds biggest companies suddenly need electricity the way they once need land, and there is none, nowhere near enough of it, not for what they're building.

The hyperscalers alone, Amazon, Google, Meta, Microsoft, are collectively expected to spend somewhere in the region of $690 billion on AI data center infrastructure in 2026.

Here is the thing that keep getting lost in all the coverage of chips and models and spectacular soaring valuations, which is that none of this compute works without the electricity to power it, and the electricity is the problem nobody solved.

Microsoft has described an $80 billion Azure backlog driven not by weak demand but by power that isn't connected yet.

We must secure the power, we must build the grid, we must find the fuel that can run a data center at 3am on a Tuesday when the wind has stopped and the sun won't rise for six more hours, the kind of power that is patient and dense and frankly and without sentimentality available.

The grid has its own dark arithmetic. The electrons do not care about your valuation.

X-energy builds something called the Xe-100, a pebbled, pressured and pretty peculiarly designed high-temperature gas-cooled reactor, helium-cooled, fueled by billiard-ball-sized pellets of uranium.

80 megawatts per unit!

Amazon has committed to deploying more than 5 gigawatts of these reactors by 2039, which is what convinced institutional investors to mob the offering, and the customer pipeline for X-energy's power has reportedly doubled to 45 gigawatts in 18 months, a figure I keep coming back to because, look, and I don't want to alarm anyone, but that's a LOT of committed electricity from a company that hasn't yet built a single commercial unit, and either the belief is absolutely warranted or something strange is being price into the future here that we won't fully understand until its to late!

45 gigawatts. Your man running the roadshow must of had a good week.

The investment angle is this, and it bears saying plainly without too much dressed up decoration: the AI trade has a physical layer that most people haven't price yet. The shouting and shuffling and fever-pitched frenzy of frontier lab valuations absorbed all the oxygen, and while that happen, power generation infrastructure sat their patient and serious and stacking customer commitments.

There is a verse in all of this, if you follow it: power and price and patience entwined, a grid that hums for the machine that learns to find, the fuel that burns and the capital that binds, and yes the meter there is slightly of but so is the whole situation.

I will be straight about the risk:X-energy's stock is down roughly 38 percent from its opening-day surge. Fervo, the geothermal company that IPO'd in May 2026, is also down. Going public is not the same as going profitable.

Building nuclear reactors still requires regulatory clearance, construction discipline, and execution that very few companies have ever managed at scale.

The pebble-bed reactor that was once PowerPoint vaporware just got 15 times oversubscribed. Something has shifted.

The power grid is no longer boring!