On September 16, eighteen companies that routinely compete with one another announced a shared rulebook for how their data centres will behave when the grid is under stress. The telling detail sat underneath the announcement rather than inside it: a set of obligations — ride-through, curtailment, contingency response — that every member agrees to accept before it is allowed to connect. After two years of the largest data-centre operators asking their utilities for more power, several of them have discovered that the fastest route to capacity runs through giving up some control over when they consume it.
The AI Energy Management Alliance is the industry's answer to that deadline — eighteen members pre-agreeing the technical obligations so a flexible load becomes a service rather than an exception.
The binding constraint is not engineering. It is who pays for the transmission upgrades that make flexibility physically possible.
A queue became a rulebook
Interconnection queues have always been an engineering problem with a political coating. That balance shifted in October 2025, when the Secretary of Energy used section 403 of the Department of Energy Organization Act to direct FERC to consider a national approach to connecting large loads, opening docket RM26-4-000. FERC issued an advance notice of proposed rulemaking a few weeks later and told reporters in April 2026 that it would act by June.
What arrived on June 18 was more concrete than a proposed rule. FERC issued tailored show cause orders under section 206 of the Federal Power Act to all six jurisdictional operators: PJM (EL26-67-000), MISO (EL26-70-000), SPP, CAISO, ISO New England and NYISO (EL26-69-000). The commission preliminarily found that existing tariffs may be unjust and unreasonable precisely because they were written for loads that do not vary.
Tailored show cause orders
PJM, MISO, SPP, CAISO, ISO New England and NYISO were each told to justify their tariffs or file changes within 60 days — by August 17, 2026 · FERC, 2026
The obligations attached to those orders are unusually specific. Each operator had to file an informational report on generation adequacy for existing and prospective large loads within 30 days; those reports arrived on July 20. Requests for abeyance were due within 45 days and can pause a proceeding for up to 90 days while stakeholders build their own proposals. Parties then have 30 days to respond to whatever gets filed.
| Milestone | Date | What it settled |
|---|---|---|
| DOE §403 direction | Oct 23, 2025 | Docket RM26-4-000 opened on large-load interconnection |
| PJM co-location order | Dec 2025 | Transparent rules required for substantial loads sited with generation |
| SPP HILL initiative | Jan 2026 | Fast-track protocols for high-impact large loads approved |
| Section 206 orders | Jun 18, 2026 | Six operators told to justify or rewrite tariffs; five reform areas floated |
| Adequacy reports filed | Jul 20, 2026 | Generation-adequacy picture for large loads enters the record |
| Tariff filings due | Aug 17, 2026 | The 60-day deadline; party responses due 30 days after filing |
| Ratepayer Protection Act | Sep 17, 2026 | House passes a bill encouraging states to write large-load tariffs |
| Senate bill introduced | Sep 18, 2026 | 18-month statutory mandate, 100 MW threshold, hybrid facilities at one point |
Docket milestones compiled from FERC news releases and docketsheet entries, 2025–2026
The industry answered with a trade association
The AI Energy Management Alliance launched on September 16 with eighteen founding members, convened by Emerald AI, Google and NVIDIA and including Anthropic, National Grid, AES and NRG Energy. Its stated purpose is narrower than the promotional framing suggests: define what a flexible data centre owes the grid before the interconnection request is filed, so that ride-through, curtailment and contingency-response duties are settled in advance rather than litigated one queue at a time.
Two weeks ago we covered the alliance's launch and the 100 GW figure attached to it, and flagged that the number depends on market rules that did not yet exist. As we wrote in September, the demonstrations are real and narrow. What has changed since is that the rules are no longer hypothetical: FERC has a docket, the operators have filed, and Congress has started writing.
There is a cost to “shared governance” ISO gridlock that shows up in a tariff filing: the physical grid does not wait for administrative consensus to finish forming.— Mica C. Tufillaro, engineering analyst (RTO Insider, September 27, 2026)
AEMA's interconnection claim
The alliance argues that making demand response integral to data-centre development could allow an extra 100 GW to connect. A 2025 Goldman Sachs study cited in the launch coverage put the equivalent figure near 76 GW · AEMA and TechCrunch, 2026
The ratepayer objection is structural
None of this is disputed by the participants. The dispute is about the bill. Flexibility is cheap to require and expensive to enable: the transmission upgrades, dynamic operating envelopes and settlement rules that let a data centre shed load on command are paid for by someone, and the default answer in every RTO tariff is the residential ratepayer.
That argument became concrete in the second half of September. MISO stakeholders repeated demands for fair cost splits on large-load transmission work on September 20. On September 23, FERC commissioners exchanged concurrences on a data-centre contract dispute, declining to decide it but arguing for clearer terms and highlighting what one of them called the paucity of a $1 letter of credit standing behind a large-load interconnection deal. On September 25, Texas tightened its de facto pause on data-centre interconnections with ERCOT. The reliability regulators met the same week to discuss NERC's plans for standards development on large loads.
The measured effect is modest next to the advocacy numbers. Work out of Duke University's Nicholas Institute put the realistic contribution of data-centre demand flexibility at a 1–2% reduction in system peak, worth roughly 0.5–2.8% lower rates. Two years of Google demonstrations have produced a national demand-response portfolio of about 1 GW, and the most-cited single project is a Virginia facility of nearly 100 MW. None of that is trivial at grid scale. It is also not 100 GW.
What a flexible load actually has to accept
Curtailment: a dispatch obligation with a defined notice period and a settlement price, so the operator knows in advance what a call is worth.
Contingency response: the ability to move load within seconds of a disturbance — IEEE panels describe dynamic operating envelopes that report a few hundred megawatts of headroom inside a minute, replacing the static worst-case hosting-capacity promise.
None of these are software features. Each is a term sheet.
Will a federal flexible-load standard exist before the end of 2027?
Probability: 60% — MISO has already filed its reliability requirements tariff, the Senate bill would impose an 18-month deadline, and every operator now has a live 206 docket pushing it toward a filing.
✅ Arguments for
AEMA gives operators a ready-made technical standard, which lowers the cost of saying yes.
Confirmation criteria: FERC accepts or rehears a flexibility tariff in at least three of the six dockets; the Senate bill clears committee.
❌ Arguments against
State-level resistance is hardening — Texas and ERCOT have effectively paused new data-centre interconnections.
Refutation criteria: tariffs land but with no dispatch obligation attached, or FERC defers the whole question to a later NERC standards process.
FERC orders on the six EL26 dockets — an accepted or rejected tariff is the clearest read on where the commission lands
NERC's large-load standards project, which could absorb the technical layer the RTOs are inventing in parallel
Whether the Senate bill's 100 MW threshold survives committee, since that is the line below which most commercial data centres sit
Any published interconnection agreement that names a settlement price for curtailment — that is where advocacy numbers become revenue
Development scenarios
🟢 Standardisation (40%)
Consequences: interconnection timelines for flexible sites compress measurably, and the alliance becomes the standards body the industry actually cites.
🟡 Fragmented patchwork (45%)
Consequences: most likely path. AEMA keeps convening, states keep legislating, and the 100 GW claim stays a talking point rather than a queue item.
🔴 Regulatory retrenchment (15%)
Consequences: interconnection queues stay slow, and the alliance's pre-agreed obligations have no buyer.