FERC Just Put AI's Power Problem on a 60-Day Clock

The constraint on the next phase of AI was never going to be a clever architecture or a faster chip. It is electricity, and the machinery that moves it. On June 18, 2026, the Federal Energy Regulatory Commission made that official, issuing six tailored show-cause orders to every major US grid operator and putting the country's data-center power crunch on a regulatory clock measured in weeks.

What FERC actually did

The orders went to PJM, MISO, SPP, CAISO, ISO New England, and NYISO — the regional transmission organizations that run the wholesale grid for most of the United States — along with their transmission owners. The instruction is blunt: within 60 days, each operator must either explain why its current tariff remains "just and reasonable" without specific provisions for large loads, or propose the revisions to fix it. FERC is not asking whether the rules need updating. It is asking the operators to prove they don't, knowing they can't.

Layered on top is a faster deadline. Within 30 days, each operator must file a resource-adequacy report detailing how it will keep enough generation online to serve both existing customers and the wave of new large loads lining up to connect. FERC named five categories of reform it expects every response to address, from how co-located generation is treated to a more efficient transmission-study process. This is a defend-or-revise ultimatum aimed at the slowest, least glamorous part of the entire AI supply chain: the interconnection queue.

Why the queue is the story

For two years the AI narrative has been about gigawatts as a flex — the bigger the cluster, the bigger the headline. What those announcements quietly omit is that a signed power-purchase agreement and an energized substation are different things, often years apart. A hyperscaler can commit billions to a campus in a quarter; getting that campus interconnected to the grid can take the better part of a decade under the rules written for a slower era. The models keep getting hungrier on a software clock. The grid responds on a permitting clock. The gap between those two clocks is where the buildout actually lives or dies.

That gap is what FERC just reached into. Large loads — a single AI data center can now request more power than a mid-sized city — were never contemplated by tariffs designed around incremental industrial growth. The result has been a patchwork: operators improvising rules for co-location, behind-the-meter generation, and cost allocation case by case, with no consistency and plenty of litigation. The June 18 orders are an attempt to force a coherent framework into existence before the improvisation breaks something.

The two ways this goes

The optimistic read is that this is exactly the unglamorous plumbing work the AI economy needs. Clear, consistent interconnection rules let capital flow toward the projects that can actually be powered, instead of toward whichever developer is best at gaming a queue. Faster study processes mean a data center that can demonstrate it won't destabilize the grid gets connected in a reasonable timeframe rather than waiting behind a thousand speculative requests. Done well, standardization is an accelerant.

The cautionary read is who pays. Every reform category FERC flagged is, underneath, a fight over cost allocation — when a single customer triggers the need for new transmission or new generation, does that customer pay for it, or does the cost smear across every household on the system? The commission explicitly framed the orders around protecting existing consumers from subsidizing the buildout. But "protect consumers" and "don't slow down the data centers" are in tension, and the 60-day filings are where that tension gets adjudicated. The companies racing to build will push for socialized grid costs and expedited connection. Ratepayer advocates will push back. FERC just scheduled the collision.

The buildout's externalities keep surfacing

It is worth noticing the pattern. A few days before these orders, the Department of Justice was in court arguing that an AI company's unpermitted gas turbines were a matter of national security. Now FERC is forcing six grid operators to rewrite their rulebooks because the existing ones can't absorb data-center demand. Different agencies, same underlying fact: the physical cost of training and serving frontier models has grown large enough that it can no longer be kept off the slide deck. Power, water, transmission, emissions — the externalities the compute race prefers not to discuss are being dragged, one by one, into the open by the institutions that govern them.

That is what makes this a story for the AI beat and not just the energy desk. The pace of model progress is increasingly downstream of decisions made in interconnection proceedings most technologists have never read. Whether the next cluster comes online in 2027 or 2030 may hinge less on a research breakthrough than on how PJM answers a FERC filing this summer.

What to watch

The honest framing is that nothing is resolved yet. Show-cause orders open a process; they do not dictate the outcome. The operators will file, intervenors will respond, and the real shape of the new rules won't be visible for months. Some operators may defend their tariffs and pick a fight; others will propose sweeping rewrites. The five reform categories are the map of where the leverage sits.

But the signal is unambiguous. The federal regulator that governs the grid has decided the AI buildout's power demand is no longer something the system can muddle through. For an industry that likes to measure progress in benchmark points and token counts, the most consequential number this quarter might be 60 — the number of days six grid operators now have to explain how they plan to keep the lights on for everyone else while the data centers arrive.