For two years the AI shortage story was about silicon: chips first, then the memory beside them. That story is now half-finished. The part increasingly deciding how much AI can actually be built is more basic — electricity, and the grid that carries it.
A data center is a power plant with servers attached
A modern AI cluster draws as much power as a small city, continuously. Getting that power isn’t a matter of ordering more — it means substations, transformers, transmission upgrades and grid-interconnection queues measured in years. You can add chips in months; you cannot add a gigawatt of firm power on the same schedule.
Why the constraint is moving here
As chip and memory supply catches up, the binding constraint migrates to whatever is slowest next — and right now that’s the physical delivery of electricity: grid capacity, the power-delivery chain into the rack, and the equipment that makes both possible. When a constraint is scarce and slow to expand, the companies that relieve it gain pricing power.
Who sits on it
The names exposed aren’t the AI labs — they’re the less glamorous providers of power and electrical equipment: grid gear, on-site power, cooling, and the utilities feeding the load. THE ENTITY maps each of these to the exact part of the chain it relieves, so you can see the exposure rather than guess at it. Research, not advice.
Go deeper in the model
See it live on the map
Plans from $19/mo — cancel anytime. Two minutes to your first read.