There is a widespread assumption that the constraint on building new electricity generation in the United States is money, or public opposition, or the physical difficulty of construction. For a large share of projects, it is none of those. It is a queue.
Before any generator can deliver power, it must be studied: the operator has to determine what upgrades the grid needs to accept it without destabilising. That study process was built for an era of a few large plants a year. It is now receiving applications by the thousand, and the backlog of capacity waiting in it exceeds the capacity currently operating on the system.
Why the line does not move
The mechanics are almost self-defeating. Studies are conducted serially and each one assumes a specific set of projects ahead of it in the queue. When a project withdraws, which happens constantly because developers file speculative applications at multiple points to preserve optionality, the assumptions change and studies must be redone. The queue restudies itself faster than it clears.
The costs allocated by those studies are the other half of the problem. A project can wait years only to be assigned network upgrade costs large enough to kill it, at which point it withdraws and every project behind it is restudied again. Developers describe filing several applications for a single intended project precisely because the process is unpredictable, which inflates the queue further and makes it slower for everyone, including themselves.
The reforms underway are sensible and slow. Cluster studies assess groups of projects together rather than serially. Deposits and readiness requirements filter speculative applications by making them expensive. First-ready rather than first-come ordering advances projects with land, permits and offtake secured. Where these have been implemented, throughput has improved measurably, and the improvement is measured against a backlog so large that improvement and resolution are different things.
Demand is not waiting for any of it. The load growth forecasts that utilities revised upward, driven substantially by the data centre construction behind the AI infrastructure buildout, assume generation arrives on a schedule the queue cannot currently support. Several utilities have responded by extending the life of existing plants they had planned to retire, which is the clearest possible signal about what they expect the queue to deliver and when.
Transmission is the deeper constraint underneath the queue, and it is worse. New generation is frequently in places the existing network was not built to serve, and building the lines to reach it runs into siting processes spanning multiple states and jurisdictions, each with a veto and none with responsibility for the whole. That is the same administrative capacity gap now recognised as the real obstacle across infrastructure generally, and it has the same shape here: the money exists, the projects exist, and the process connecting them was designed for a different volume of work.
The comparison worth drawing is to the rural broadband buildout, where the expensive part turned out to be the last stretch rather than the first, and to the replacement cycle now reaching small water systems. In each case a funded programme met an institution without the staff to execute it, and the shortfall showed up as delay rather than as a visible failure.
What makes the interconnection queue different is that its delay is measurable to the megawatt, and it is being measured. That is unusual in infrastructure policy, where problems normally have to become emergencies before they acquire a number. Here the number exists, it is published, and it has been growing for years.
