Grid capacity

Grid capacity is the amount of electricity the local power grid can actually deliver to a data center. It matters because everything stacked above it — chips, fabrics, models — is inert without delivered power, and grid timelines dwarf both chip and building timelines. Right now AI training and inference loads are requesting power at a pace utilities simply cannot match, which turns grid access into the binding constraint on new capacity: the wafers and racks may exist, but the electrons to run them do not.

If access stays tight, interconnection queues stretch out for years, stranding built shells and forcing curtailment or behind-the-meter generation. That routes value to whoever can build or supply power — engineering contractors like Quanta Services, merchant generators such as Talen Energy and Vistra, grid-equipment makers like Hitachi Energy, and newer site-and-power players including Fermi and small-reactor developer Oklo, all of whom sell into a constraint that clears first and governs everything downstream.