Data centers are building power behind the fence. It won’t stay there forever.

Large data center facilities are increasingly building their own power. Not because they want to run power plants, but because in many markets the grid simply cannot connect them on the timelines these projects require. Interconnection queues stretch years into the future, and in some locations, only limited capacity is available. For developers trying to move at the pace of AI infrastructure, self-supply becomes the only practical path forward.

In many cases, the result is a hybrid architecture. A limited grid connection provides part of the supply, while substantial on-site generation closes both the capacity and reliability gap. The grid supplies energy when available; the on-site plant ensures the facility can operate at full load with the level of reliability required for these operations. The facility is no longer just a customer of the grid; it also contains generation connected to the system.

That changes how the system behaves. Power flows shift. Local system conditions can change. Generation may now sit behind what was previously modeled simply as load. If that generation is configured to export, it can eventually operate as a resource for the broader system as well.
Some projects go even further, operating as fully self-supplied facilities. In those cases, reliability must be built entirely within the fence line. Redundant generation, fuel security, spinning reserves, and contingency capability are all part of the project to approximate the reliability the grid normally provides.

Many of these facilities still expect to expand their grid connection over time, because the grid remains the most efficient way to deliver very high levels of reliability. When that happens, the generation built inside the fence line can become part of the broader system, adding operational flexibility and strengthening reliability for the surrounding network. For utilities and regulators, the timing is difficult to reconcile.

Transmission and grid infrastructure take years to study, permit, and build. Projects that eventually seek a grid connection still have to enter the interconnection queue and wait for the system upgrades required to support them. The constraints that pushed developers toward self-supply don’t disappear simply because a project later decides to connect. And when those facilities do connect, the generation built inside the fence line doesn’t disappear. It becomes part of the system.

That raises a question for planners. If large loads increasingly arrive with their own generation, should the system continue to treat each connection purely as a new demand point, or should planning begin to account for the combined resources that accompany them?

The opportunity is to design the grid so that, when these facilities arrive, the resources they bring strengthen the system rather than strain it.