You cannot plug a datacenter into an application. This should be the easy part of the AI infrastructure discussion, but the applications have become impressively large.

Axios reports roughly 2,600 gigawatts of proposed generation and storage awaiting U.S. grid connections; many projects may never materialize. That last clause belongs beside the number, not somewhere beneath the celebration.

Follow the mechanism. A proposed project enters the connection process. It still needs to become an operating project connected to a functioning system. Until that happens, its presence in the queue establishes that somebody wants to build something. It does not establish that electricity is available to anybody.

Make powerful people uncomfortable.
Subscribe to Democrawonk.

Adding up proposals is useful planning information. Promoting the sum into a working future would be like counting everyone waiting outside a restaurant as dinner already served. The queue may demonstrate enthusiasm. Try eating it.

This is different from the financing risk of the AI buildout. Even a well-funded proposal must become a physical connection. A confident investor cannot authorize electrons to skip the unfinished section.

There are real projects and useful technology here. The objection is to treating proposed capacity as an accomplishment before it exists. A list can grow faster than an electrical system. If the presentation cannot tell those two things apart, its most advanced feature is the font.