Power Bottleneck: AI Growth Hinges on Global Electricity Grid
A grid constraint behaves differently from a price rise. When power is simply unavailable in a region, capacity stops being something you can buy your way out of and becomes something you schedule around — which is a planning problem long before it is a budget problem.
For teams running agents, rationed compute makes wasted work visible in a way cheap compute never did. Two agents solving the same task, a loop nobody noticed, an expensive step that should have been gated behind a human — all of it is affordable when capacity is elastic and none of it is when capacity is queued. The operational discipline that looks optional today is the thing that decides how much you get done per available kilowatt tomorrow.
Global data center power use to hit 945 TWh by 2030 as grid capacity—not GPUs—limits AI growth. Ohio, Dublin and PJM impose moratoriums; nuclear deals surge.
Read the full report at informedclearly.com →Pragor's editorial analysis. Underlying reporting by informedclearly.com, our media partner. Quotes and facts belong to the original publisher.