FERCの大規模負荷命令：米国の6つの系統運用機関が答えるべきこと

Most of the argument about AI and electricity is conducted in forecasts. The part that will actually decide where the next few gigawatts of computing get built is duller than a forecast and much more specific: the tariff language that says how a very large customer gets connected to a grid, who studies it, and who pays for the wires.

In the United States that language is being rewritten right now, under a deadline, in six places at once.

What FERC did on 18 June

The Federal Energy Regulatory Commission issued show cause orders under Section 206 of the Federal Power Act to every one of the six FERC-jurisdictional grid operators: PJM in docket EL26-67, SPP in EL26-68, NYISO in EL26-69, MISO in EL26-70, CAISO in EL26-71 and ISO-NE in EL26-72 (Husch Blackwell, on the 18 June 2026 orders).

A show cause order is not a rule. It is an instruction to prove that the rules you already have are just and reasonable, with the clear implication that if you cannot, you will be filing new ones. Each operator had to demonstrate “whether its existing tariff provisions adequately enable” the efficient and reliable interconnection of large loads, and if not, to propose revisions under Section 205 (Bracewell, 9 July 2026).

A large load, in the definition FERC put forward, is a new commercial or industrial customer at a single site, with peak demand of 50 MW or more, interconnecting at transmission-level voltage above 69 kV, and not part of a co-location arrangement. Operators may propose their own definition instead, but have to show it is just and reasonable; FERC accepted SPP’s, which reaches down to 10 MW at or below 69 kV (Bracewell; RMI).

A co-located load is “an end-use customer physically connected to a generating facility on the customer’s side of the point of interconnection” (Bracewell). The data centre and the power plant are on the same side of the meter. That is the arrangement that has made the last two years of US energy law interesting, because a customer on the plant’s side of the point of interconnection is, arguably, not using the transmission system at all, and therefore not paying for it.

The five things each operator has to address

FERC named five reform areas (Husch Blackwell; RMI):

1. Efficient transmission service application and study processes for large loads.
2. Cost transparency, so that a new large load does not shift costs onto existing transmission customers.
3. Rules and rates for co-location agreements and behind-the-meter generation.
4. New transmission services for flexible large loads.
5. A process for studying generating facilities that serve electrically proximate and co-located large loads.

PJM was not ordered to address co-location, because it was already doing so in its own docket, EL25-49, and SPP was excused from the standard service agreement and proximate-generation items for the same reason, having dealt with them in earlier proceedings (RMI).

ERCOT and the regions where transmission utilities never ceded operational control to an RTO or ISO are outside these orders entirely (Bracewell). Texas, in other words, is not part of this. Neither is most of the Southeast. FERC has encouraged non-RTO utilities to file similar revisions but has not ordered them to (RMI).

The calendar, and why mid-November matters

The orders carried a ladder of dates: motions to intervene by 9 July, an informational report on resource adequacy for existing and new large loads by 20 July, show cause responses or Section 205 filings by 17 August, and comments from interested parties thirty days after that (Husch Blackwell).

There was also an exit from the August deadline. An operator could request, by 3 August, that its proceeding be held in abeyance for up to 90 further days, non-extendable, with the request describing the Section 205 filing it intended to make and when (Husch Blackwell). For a proceeding that took that route, the filing is due 15 November and responses from other parties a month later (RMI).

We are not naming which operators took the abeyance, because we have not verified it in the dockets themselves, and an unofficial mirror of a FERC docket is not a source. The structure is public either way, and it puts the documents that will set the terms for connecting a 50 MW-plus customer to most of the American grid in this month and next, rather than in some future rulemaking.

What is actually being decided

Who pays for the wires. A large load arriving at a substation can require network upgrades that outlast it. If those costs land in the general transmission rate, every existing customer in the zone pays for a facility built for one company. Reform area two exists because that allocation has, so far, been made case by case.

Whether you can opt out of the grid while standing next to it. Co-location is the cleanest way to put a gigawatt of demand next to a gigawatt of supply with no new transmission at all. It is also the cleanest way to remove a large generator from the pool that serves everyone else. FERC’s definition pins the question to a precise physical fact, which side of the point of interconnection the customer sits on. An engineer can answer that from a drawing.

What flexibility is worth. Reform area four, “new transmission services for flexible large loads”, is the one with the most room in it. A data centre that can be curtailed is a fundamentally different object, from a planner’s point of view, than one that cannot, and the US grid currently has almost no tariff vocabulary for the difference. Whatever language appears in these filings will be the first attempt at pricing it at scale.

The same problem, three different instruments

Japan, the EU and the US have each arrived at the collision between AI compute and electricity, and each has reached for a different tool.

Japan’s binding constraint is physical. The question there is not how a large load is studied but whether the interconnection capacity exists to move power from where it is generated to where the fabs and data centres are going, which is why the live argument is about undersea HVDC and the queue, not about tariff definitions (our briefing on the Hokkaido data centre cluster and the HVDC link; on the storage interconnection queue).

The EU’s instrument is disclosure. Rather than rewriting connection rules, it has started requiring data centres above a size threshold to report energy, water and waste heat, on the theory that measurement precedes regulation (our briefing on the EU data centre reporting rules).

The US is doing neither. It is rewriting the contract between a very large customer and the system, one market at a time, on a six-month clock. That is the fastest of the three instruments and the least uniform. Six operators answering the same five questions will not produce one answer, and the differences between them will show up as siting decisions within a year.

Three facts decide a project’s timeline in this environment: which definition of “large load” applies in the market it is sited in, whether the arrangement is co-located under FERC’s physical test, and what that operator filed.

At the Summit

Tech for Impact Summit 2027 is co-hosted by the United Nations University and held at its headquarters in Shibuya, Tokyo on 18 and 19 May 2027, and the Tech for Impact series runs across Asia, Europe and the US. The energy and infrastructure sessions put grid planners, the companies siting compute, and the capital behind both at the same table, precisely because the three rarely read each other’s documents.

The Summit is invitation-only. If you work on interconnection, data centre siting or the financing of either, you can join the waitlist at tech4impactsummit.com/apply and we will be in touch.