Why AI Data Centers Are Being Built Near Power Plants

The largest AI training campuses announced in 2024–2026 share a striking pattern: they are being sited next to existing power plants, often with direct connections that bypass the public grid.

Last updated: 2026-04-15
Reading time: 6 min

The transmission queue problem

Connecting a new gigawatt-scale customer to the existing high-voltage grid can take five to eight years or more in the most congested regions, and PJM's queue has averaged well over eight years from request to commercial operation. AI operators racing to deploy GPU clusters cannot wait that long.

Siting next to an existing nuclear, natural gas, or large solar+storage facility lets the operator energize in 12–24 months instead of years.

Behind-the-meter explained

A 'behind-the-meter' configuration means the data center buys power directly from the generator without it flowing through the regional transmission system. This avoids transmission upgrade costs and queue delays — but it also raises regulatory questions about cost-shifting and grid reliability.

Where this is happening

Pennsylvania, Texas, Louisiana, and parts of the Midwest have seen the most activity. Susquehanna (Pennsylvania) and several ERCOT-region sites are well-known examples.

Frequently asked questions

Is behind-the-meter siting legal?

It is allowed in many states, but federal regulators (FERC) and state commissions are actively reviewing how to handle very large behind-the-meter loads.

Does this make the grid less reliable for everyone else?

It can — if a large generator is dedicated to a single data center, less capacity is available to the broader grid during peak hours. Regulators are studying this trade-off.