Power, Not Silicon: What FERC's Large-Load Order Means for Where AI Compute Lands
The binding constraint on AI is quietly shifting from chips to power — and to the grid connection that feeds them. On June 18, 2026, FERC made that shift official policy.
What FERC did
The Federal Energy Regulatory Commission issued tailored show-cause orders under Section 206 of the Federal Power Act to all six grid operators it oversees — PJM, MISO, SPP, CAISO, ISO-NE, and NYISO — directing each to justify or reform the tariff rules that govern how data centers, manufacturing facilities, and other large energy users connect to the grid. Each operator has 60 days to defend or revise its tariffs, and 30 days to file a report on how it will ensure adequate generation for existing and new large loads.
The orders tee up five categories of reform:
- Faster transmission-service application and study processes, including alternative transmission technologies
- Preventing cost-shifting onto other ratepayers, with transparency into transmission costs
- Accommodating co-location agreements and behind-the-meter generation
- Providing new transmission services for flexible large loads
- A process to study generation that serves electrically proximate and co-located loads
FERC frames the action as delivering "speed to power" for the innovation economy and the global AI race, balanced against consumer protection. It builds on a December 2025 order directing PJM to adopt transparent co-location rules and on SPP's High Impact Large Load initiative, following review of more than 3,500 pages of public comments.
Why it lands on AI compute
"Large load" is regulatory speak for what this industry calls a data center. The economics of AI training and inference increasingly hinge less on a GPU's list price and more on whether — and how fast — you can energize the megawatts to run it. Interconnection queues, transmission studies, and tariff terms now sit squarely on the critical path of every large build.
Two phrases in the order matter most for compute planners:
- Co-location and behind-the-meter generation. This is the formal recognition of what neoclouds and hyperscalers are already doing — siting capacity next to power (gas, nuclear, renewables) to skip the queue. The rules for that are now in play.
- Cost-shifting and transmission-cost transparency. How grid costs get allocated to large loads will flow downstream into the effective cost of compute — and it will differ by region.
The part that matters for where workloads run
The most consequential detail is that FERC explicitly declined a one-size-fits-all fix. Six operators, six market designs, six different answers — and at least 60 days of divergence ahead. So, the map of where you can stand up AI capacity quickly, and at what "all-in" cost, is about to become more regionally fragmented, not less — especially as state and local siting rules combine with these new federal requirements.
That is the same question we ask at AIForge Works from the price side: where can this workload actually run, and what will it cost across providers and regions? The grid is now an input to that answer. The cross-region differences we already observe in public compute pricing are, in part, downstream of power economics and interconnection rules — and FERC's action will reshape both, region by region, over the coming months.
We don't take a position on policy; we track what's published and normalize it. But this is a clean marker worth watching: the binding constraint on AI compute is moving to power and the grid, and where is becoming as decisive as how much.
Source: FERC, "FERC Launches Aggressive Targeted Action to Speed Large Load Integration," June 18, 2026. Informational only; not legal, regulatory, or investment advice.