The Number

On September 1, 2026, a mapping exercise circulated showing every AI mega-campus of one gigawatt or more currently operating, building, or announced in the United States. Thirteen sites. Roughly 75 gigawatts of announced capacity.

For scale: the entire United States nuclear fleet is 94 reactors producing roughly 97 gigawatts, assembled over five decades of national effort. Thirteen private campuses, announced in about three years, represent something close to 77 percent of that.

The named sites tell you where the century is being built: Matador and HyperGrid at 11 GW, Stratos at 9, Nexus Hubbard at 7.2, Meta Hyperion at 5, Data City at 5, Homer City at 4.5, Delta at 4, Fairwater at 3.3, Camelia at 3.2, AWS Rainier at 2.2, xAI Colossus at 2, PORTS in Ohio, and the rest.

A note on the source, because it matters. That map was produced by an AI tool in a paid partnership and compiled from announcements rather than from interconnection agreements. Announced capacity is not energized capacity, and a meaningful share of announced projects never energize at the stated number. Treat 75 GW as the ceiling of what has been claimed, not a measurement of what exists. The direction of the argument does not depend on the precision of that figure.

The Fleet Ceiling Was Never Theoretical

In The Role of Energy I argued that the size of a national autonomous workforce is capped by generation, not by ambition. A humanoid draws roughly 1.125 kilowatts. Multiply by the fleet and you get a hard number the grid either supplies or does not. I called it the fleet ceiling. It was, at the time, an argument.

It is no longer an argument. It is a map with thirteen circles on it, and every one of those circles was placed by an engineer asking a utility a question about capacity.

Notice what the map is actually showing. It is not a map of talent. It is not a map of capital, which is fungible and can be wired anywhere in an afternoon. It is not a map of demand, which is national. It is a map of where the electrons are. Every circle sits where a utility could answer yes.

This is the clearest confirmation yet of the proposition at the center of this framework. Kilowatts are the new man-hours. Not as metaphor. As a conversion rate that decides which places participate in the machine economy and which places watch it.

Compute First, Then the Machines

Seventy-five gigawatts of announced compute is not the end of the demand curve. It is the first order of it.

Compute is the cheap load to site, because a data center is a building with a substation. The harder load arrives next, and it is the one this framework has been counting all along: the embodied fleet. Robots on factory floors, in warehouses, on construction sites, in eldercare. Each drawing its 1.125 kilowatts, continuously, in places where people actually live and work rather than in a fenced campus off an interstate.

If thirteen campuses can claim 75 GW, ask what a national autonomous workforce claims at scale. Run the arithmetic in the fleet ceiling and the answer is that the compute buildout is the smaller half of the problem. We are provisioning for the racks. We are not provisioning for the arms.

The policy conclusion has not changed and this map sharpens it. An Apollo program for the grid. Ten to twenty gigawatts of additional capacity this decade provisioned specifically with industrial automation in mind, not as a residual of data center planning. Robotics treated as a planned major load in electric resource planning, the way compute finally is. Corridors where energy allocation, land and permitting are pre-cleared.

Where the Circles Land, and Why It Is Not the Coasts

Look at the geography. The circles cluster in Texas, the Ohio Valley, the Southeast and the Gulf. They are not in San Francisco, Boston, or Seattle. The regions that spent forty years being told their industrial era was over are the regions being selected to host the next one, for the same reason they hosted the last one. Electricity, land, and basic infrastructure.

In Revitalize America I made this case for the Rust Belt and called it the Robot Belt. The map obliges me to extend it south. Louisiana, Mississippi, Alabama, Tennessee and the Ohio Valley are absorbing a buildout that the coastal press still reports as a technology story rather than what it is, which is the largest industrial construction program in the United States since the war.

Meta has committed tens of billions in Richland Parish, Louisiana. Applied Digital has a multi-billion dollar campus in Rapides Parish. Hut 8 is at River Bend. Gas plants are moving through the state commission to feed them. The Gulf has the cheapest industrial power in the nation, the gas basin underneath it, rivers that move heavy freight, and parishes where land trades in the low thousands per acre with no zoning board standing between a builder and a permit.

That is not a coincidence of siting. It is the fleet ceiling expressing itself geographically. Capacity goes where the grid says yes, and the grid says yes where somebody already built generation and nobody has built a veto.

The Question the Map Does Not Answer

Seventy-five gigawatts will be built by somebody. Thousands of electricians, pipefitters, ironworkers and operators per campus, for three to five years each, in counties and parishes whose entire labor force is smaller than the crew. The buildout is a labor event before it is an automation event, and almost nobody is modeling it that way.

And when the campuses are finished, they will employ a hundred people each. That is the arithmetic that will decide whether these regions remember this decade as the one that brought them back or the one that used them. It is the same arithmetic that produced Youngstown.

The answer is not to slow the buildout. It is to make sure the hours it creates are counted, and that the people who live where the circles land hold a claim on them. Who owns the robots is the question. The map just told us where they are going to stand.

Seventy-five gigawatts is not a technology headline. It is a national industrial mobilization happening without a national industrial strategy. We built the Interstate Highway System with one. We built the TVA with one. We are building this one campus at a time, negotiated privately, parish by parish, and calling it a market.

Nations run on hours, not just dollars. Count the hours. Provision the grid. And make sure the working man is on the deed.