FERC's large-load orders moved AI power demand into six RTO and ISO dockets [1][2][3][4]
The prior file at ngtimes.org/2026/06/21/ferc-large-load-orders-start-ai-power-filing-clock asked for a public receipt before the frame hardened. Today's record supplies one, but it does not settle every claim.
What changed is procedural gravity. The commission used its Federal Power Act Section 206 authority to open show-cause proceedings directing every major organized market, including PJM, MISO, CAISO, ISO-NE, NYISO, and SPP alongside ERCOT's separate state track, either to justify existing rules for integrating very large loads or to reform them. Show-cause orders are not suggestions. They start a filing clock, compel a written record, and end with commission action if answers disappoint. Six operators now owe public justifications, which converts the loudest energy debate of the AI boom into tariff text with deadlines. [1][3]
The subject matter is genuinely unsettled despite confident commentary on all sides. Data-center loads arrive with profiles grids have rarely served: gigawatt-scale demand, extreme sensitivity to interruptions, appetite for co-locating with generation behind the meter, and flexibility some operators want to harvest while others doubt exists. Existing tariffs were written for factories and substations, not server cities. Questions FERC flagged include who pays for network upgrades triggered by a single campus, whether co-located generation escapes cost allocation others bear, and how fast interconnection studies can move without cutting reliability corners. None of these has an obvious answer, which is precisely why dockets beat press releases. [1][3][4]
The MSM frame is straightforward: large-load integration is now a formal tariff and reliability issue. The X frame is sharper and less patient: AI data centers are buying their way around ordinary grid customers. Both frames miss what show-cause procedure reveals. Mainstream coverage reads the orders as technocratic housekeeping, missing that Section 206 intervention signals commissioners see market-wide problems, not regional quirks. X reads them as capture or rescue depending on the account, missing that the orders force every operator's assumptions into a public docket where advocates can subpoena the arithmetic. The paper's read is narrower: six operators now owe justifications or reforms, and filings can be audited in ways vibes cannot. [1][2][3]
The cost-allocation fight inside these dockets will touch more households than any AI story this year. Transmission upgrades for hyperscale campuses run to billions, and the difference between assigning those costs to the requesting load and spreading them across ratepayers is the difference between the AI boom funding the grid and the grid subsidizing the AI boom. Consumer advocates, industrial customers, and hyperscalers will all intervene, and their filings become the public map of who bears growth. [3][4]
Timing also matters. Interconnection delays have become the binding constraint on data-center siting, with operators publicly complaining about multi-year waits while reliability staff warn that speed without study invites blackouts. The orders put both complaints under oath, effectively. Whatever balance emerges in final rules will govern how fast AI infrastructure physically arrives. [4]
That matters because the public decision is no longer about whether the topic feels important. It is about which document controls the next claim. Here the controlling documents are the six responses, the interventions they draw, and whatever orders follow. [1][2][3][4]
The remaining gap is practical. The next receipts are operator filings, proposed cost-allocation terms, and revised interconnection timelines. Until those land, the responsible headline is a receipt check, not a victory lap. Read the dockets when they populate; the future price of a training cluster is being written there.
-- THEO KAPLAN, San Francisco