The U.S. doesn’t have a power shortage. It has a system mismatch.

The current conversation suggests we simply need more generation to support AI-driven growth. That framing is incomplete. The U.S. has fuel. It has capital. It has proven technology. What it doesn’t have is a power system designed for the kind of load that is now showing up.

Large-scale compute facilities are not traditional industrial customers. They are large, concentrated, deployed quickly, and operationally intolerant of interruption.

They ramp fast, demand five-nines reliability, and cannot sit in multi-year interconnection studies or wait on transmission upgrades. Because their load is so material — often hundreds of megawatts at a single node — they also trigger understandable regulatory and political scrutiny. Policymakers are cautious about infrastructure expansions that could shift costs to existing ratepayers if projected growth doesn’t fully materialize.
The grid, by contrast, was built for gradual expansion — regulated cost recovery, long planning horizons, predictable demand growth. That model worked when load was incremental and dispersed.

Today the friction shows up in queues, constrained substations, delayed transmission buildouts, extended review cycles, and projects paused while cost allocation debates are resolved. Capital waits. Schedules move. Approval risk becomes as relevant as engineering risk.

The mismatch is structural, and structural problems require structural solutions.