Kimmeridge Energy Management dropped a number this week that deserves more attention than it's getting: up to 50% of proposed US data centers are at risk of delays or outright cancellation. Not because AI demand is softening. Because the physical world keeps refusing to cooperate.
"The Silicon Valley model is running into a real-world infrastructure constraint," Kimmeridge managing partner Ben Dell said Wednesday. That's a polite way of saying: you can't build a hyperscale compute facility by moving fast and breaking things when the things you're breaking are transmission lines, permitting offices, and community zoning boards.
This lands on top of what I covered last week — PJM's $6 billion congestion bill for the first half of 2026 alone, a 43% surge from the year prior, driven by overloads on high-voltage transmission lines with Northern Virginia — the data center capital of the planet — among the hardest-hit areas. The grid is already buckling under demand that exists today. The demand that's still in the planning queue is what's now in question.
Here's the civilizational stakes: the IEA forecasts global electricity demand growing at 3.6% annually through 2030, with data centers as a core driver. Advanced economies are accelerating after 15 years of stagnation. We are at the inflection point. And we're about to lose half the compute infrastructure that was supposed to ride it.
The bottleneck has never been ambition or capital. It's always been the physical stack — transmission capacity, interconnection queues, local politics, grid hardware. Kimmeridge is essentially telling us that the AI buildout is going to be slower and messier than the spreadsheets assumed, not because the demand isn't real, but because megawatts don't materialize on a venture timeline.
Every delayed data center is a delayed future. The answer is transmission investment, permitting