The AI energy layer

AI compute is a machine for turning electricity into answers, and the electricity is running short before the chips do. The binding constraints in this layer are physical and slow to fix: grid capacity and the multi-year interconnection queues to plug a data centre in, the transformers and substations that step power down to the racks, the power-delivery chain inside the building, and the cooling — increasingly liquid — that keeps dense racks alive. Water availability quietly decides where the largest sites can be built at all. Because these constraints take years and heavy capital to relieve, the companies that own grid equipment, power electronics and thermal systems sit directly in the path of every new cluster — the "pick-and-shovel" names that get paid whichever model wins. This layer is where a shortage shows up first as a gating item on how fast capacity can come online.