Nuclear for AI: the gigawatt power deals, and the simpler alternative

When the people who run the world's largest computers go shopping for electricity, they are no longer calling the utility. They are buying reactors. Microsoft is restarting Three Mile Island. Meta has signed for gigawatts of nuclear. Amazon and Google are funding a new generation of small reactors. It is an extraordinary scramble, and it confesses, in billions of dollars, exactly what the real bottleneck is. The question is whether buying a power plant is the only way to answer it.

The scramble, in numbers

The deals stack up fast. Microsoft contracted for the restart of the Three Mile Island Unit 1 reactor, roughly 835 MW, on a 20-year power-purchase agreement, with the plant targeted to return around 2028 (Global Finance). Meta announced deals supporting up to 6.6 GW of nuclear capacity by 2035, including an Oklo-backed 1.2 GW campus in Ohio and a 1.1 GW agreement at Constellation's Clinton plant (Latitude Media, CBS News).

Amazon is backing up to a dozen X-energy small modular reactors and paid for data centers drawing on Pennsylvania's Susquehanna plant; Google agreed to buy power from Kairos Power's reactors, with first units expected around 2030. Across the sector, the hyperscalers have committed well over $10 billion to nuclear, with more than 20 gigawatts of projects in development (Introl).

What the reactor rush is really admitting

Read past the press releases and the message is stark. The richest companies in technology, firms that could buy almost anything, have concluded that the one thing they cannot reliably buy is electricity. Not chips. Not engineers. Power. When your answer to a shortage is to personally finance the restart of a nuclear plant, you are telling the world that the grid will not give you what you need, and that power has become the binding constraint on how big you can grow.

That is the same lesson every piece on this site keeps arriving at, now written in the most expensive ink available. The frontier is not power-hungry by accident. It is power-bound by physics. And the proof is that the smartest capital allocators on earth are spending years and billions to lock in megawatts they could not otherwise secure.

The alternative the nuclear story overlooks

Nuclear is the right instinct, abundant, dependable, around-the-clock power, applied through the slowest possible instrument. A reactor is a decade-long project wrapped in regulation. It makes sense for a hyperscaler with a 2035 horizon and a balance sheet to match. It makes no sense for an operator who needs to run GPUs this year.

And the scramble overlooks the obvious: in some places, abundant low-cost power already exists, today, no restart required. The Gulf can dedicate cheap, dependable generation to compute at a price and pace the mature grids chasing nuclear cannot match. The hyperscalers are building toward the kind of cheap, firm power that a UAE free zone can already offer. They are taking the ten-year road to a destination that, in the right geography, you can simply move into.