The Grid as Bottleneck: Why Transmission Is the Rate Limiter for AI

The binding constraint on AI infrastructure buildouts is increasingly not chip supply but grid capacity. New data center campuses need hundreds of megawatts of firm power, and getting that power to a site requires transmission upgrades and interconnection approvals that routinely take longer than the data center itself takes to build — a reversal of the usual bottleneck.

The interconnection queue is the clearest evidence of this: projects across the country are waiting years, not months, for grid operators to study and approve new connections, and the queue has grown faster than utilities' ability to process it. Transformers — a component that used to be a commodity, multi-week lead-time item — now have lead times measured in years for the largest units, because a handful of manufacturers serve the entire global market.

This bottleneck creates a specific set of winners. Grid equipment makers like GE Vernova sit directly on the critical path — turbines, transformers, and grid software all see demand pull-forward as utilities race to expand capacity. Utilities with existing generation and transmission footprints near data center demand centers, like NextEra, benefit from being able to offer capacity faster than a new entrant could build it from scratch.

The risk to this thesis is timing, not direction: grid buildouts are lumpy, and equipment makers' backlogs can convert to revenue on a slower or faster cadence than the market currently expects. GE Vernova in particular trades at a valuation that already assumes strong execution on converting its order book. We think the structural bottleneck is real and multi-year, but the individual stocks will move on quarterly conversion of backlog to revenue more than on the macro thesis itself.