A US grid operator sheds load over datacentre demand

Where things stand (as of Oct 17, 2026)

The instruction part is already met, more than once, inside the window. PJM issued a Pre-Emergency Load Management Reduction Action on Oct 13, 2026 (emergency-procedures posting 105561, all zones, 14:00–21:00) and again on Oct 14. Its Oct 14 Inside Lines update says so directly. On Sept 17 it also activated Pre-Emergency and Emergency Demand Response. Activating contracted pre-emergency or emergency demand response for an impending reliability emergency counts as an instruction under the question's definitions. PJM reports demand response was activated on 9 days in 2026 and 15 days in 2025–26, against 8 days in 2015–2024.

The attribution part is the crux. Most of PJM's event-specific documents use neutral language:

• The Sept 16 and Oct 11 applications to the Department of Energy for emergency orders cite "elevated demand, a high generation outage rate…, planned transmission outages."
• The Sept 17 and Oct 13–14 operations posts cite heat, exports and outages.
• The Oct 13 review of September speaks only of "rising overall demand."
• Duke's Sept 18 emergency application cites heat.

One document is different. On Oct 15, 2026 PJM published "Providing Reliable Electricity in a New Reality" on Inside Lines (insidelines.pjm.com/providing-reliable-electricity-in-a-new-reality/). It is a preliminary summary of the Oct 13–14 events, attributed by PA Environment Digest to PJM System Operations director David Souder. It says:

• the shoulder-season reprieve "is not what it used to be… in part because … Most of the new, rapidly growing energy demand is from data centers… a 12-month 24/7 phenomenon";
• "the result is that periods the grid once relied on for maintenance… can now become periods of significant system stress";
• "This week has been a prime example";
• demand response was activated "including Oct. 13 and 14."

Newsweek covered it on Oct 16 under the headline "Biggest Power Grid Operator Says Data Centers Leave 'Little Room for Error'."

Reading adopted. I read the criteria literally. The post names growth in data-centre consumption, explicitly, as a contributing cause of the stress condition, and ties it to the specific emergency ("this week"). The criteria allow "growth in that consumption" and do not require it to be the main cause. On that reading the question resolved YES on Oct 15, 2026.

Residual doubt. A strict judge could call this structural commentary rather than an event-specific attribution. PJM's own event-level explanation of that week stresses 61 GW of outages and warm weather, and Newsweek framed the post as a "warning." There is also some chance a resolver never finds the post. I put roughly 60–70% on acceptance once found and about 85–90% on it being found. Other chances before year-end add a little:

• PJM's revised Reliability Backstop Procurement filing, due by Oct 29;
• the Nov 5 Operating Committee review of the October event;
• the winter outlook;
• further shoulder-season events.

That gives 0.63 by 2026-12-31.

Later quarters: hazard if the October post is not accepted

Reference class. From 2025 until October 2026, despite roughly 15 demand-response days, I found no explicit event-level data-centre attribution from any operator. Operators default to "elevated demand," "weather" and "Large Loads," and naming only large loads does not count. So the hazard of an unambiguous statement is moderate, not high.

Why it should still rise:

• PJM's capacity auctions cleared short of the reliability requirement for 2027/28 (6.6 GW short) and 2028/29 (6.8 GW short).
• A study for the Pennsylvania utility commission finds PJM failing its reliability standard in 2027–2030.
• PJM's messaging has now moved to the "new reality" framing.
• FERC rejected PJM's large-load curtailment plan (the Interim Resource Adequacy Service) on Oct 9 and asked for "concrete evidence" if it is refiled. A refiling could cite specific events.
• State curtail-first rules are coming: Pennsylvania's tentative order, with a final order expected Jan 28, 2027, and Texas's 2025 large-load law (SB 6).
• Any real load shed would bring FERC/NERC inquiry reports.

I used quarterly hazards of 6–8.5% in 2027–28, tapering to 4–5.5% by 2031, with summer and shoulder seasons weighted higher. That gives about a 70% chance over 2027–31 in the strict-reading branch, and about 0.89 cumulative by end-2031. I computed this as one minus the product of (1 − hazard) over the quarters.

Strongest case against

"Event-specific" may be read narrowly. PJM's formal event documents (emergency-order applications, operations posts, committee slides) have never named data centres as a cause. Political sensitivity, an AI slowdown or new supply could keep operator language neutral for years.

Sources

• PJM Oct 15, 2026 post (above)
• PJM Oct 14 update: insidelines.pjm.com/operations-update-extends-maximum-generation-alert-calls-on-demand-response/
• Emergency-procedures posting 105561
• DOE Order 202-26-50 and PJM's Oct 11 application (energy.gov)
• PJM Sept 16 application for Order 202-26-45
• PJM Sept 17 operations update and Oct 13 September review
• PJM Summer Outlook, May 7, 2026 (general warning, before the window)
• FERC Oct 9 rejection (Utility Dive, Latitude Media)
• Pennsylvania utility commission tentative order (PA Environment Digest, Oct 16)
• Newsweek, Oct 16, 2026

Question. By the last day of each quarter, has a US grid operator issued a load-shed order or emergency curtailment that it attributes to datacentre demand?

Definitions.

• US grid operator: PJM Interconnection (PJM), the Electric Reliability Council of Texas (ERCOT), the Midcontinent Independent System Operator (MISO), or another entity acting as a reliability coordinator, balancing authority, transmission operator, or electric utility operating a public electricity grid serving US customers. An operator of a private datacentre electrical network alone does not qualify.
• Load-shed order or emergency curtailment: an operational instruction issued for a specific actual or impending grid-reliability emergency directing customers or serving utilities to reduce or disconnect electricity consumption. This includes rolling-blackout instructions, activation of contracted emergency or pre-emergency demand response, and instructions to transfer customer load from grid power to backup generation. There is no minimum duration or megawatt threshold. Issuance of the instruction counts even if subsequent developments avert physical disconnection or the instructed reduction is not achieved.
• Datacentre demand: electricity drawn from the grid by facilities housing computing, storage, or networking equipment, including their cooling and other supporting equipment. AI workloads are not required. Cryptocurrency-computing facilities count if the qualifying source identifies them as datacentres; references only to unspecified industrial or large loads do not establish datacentre demand.
• Attributes to: the issuing operator expressly identifies existing datacentre electricity consumption, growth in that consumption, or its forecast consumption during the specific emergency as a cause or contributing cause of the shortage, overload, or other reliability condition prompting the instruction. Datacentre demand need not be the sole or principal cause. The customers ordered to curtail need not themselves be datacentres.

Resolves YES if a qualifying public document establishes both the issuance of an operational instruction and the operator's event-specific causal attribution to datacentre demand. These may be established in separate documents, including a retrospective incident report; the qualifying publication date is the date on which the last necessary element becomes public.

Exclusions (resolve NO).

• General warnings about future datacentre growth, capacity shortfalls, or blackout risk without a specific issued curtailment instruction.
• Emergency declarations, watch notices, load-management alerts, authorizations to curtail, tariff changes, interconnection restrictions, and requests for government emergency powers without an actual operational instruction to reduce load.
• Merely choosing datacentres for curtailment because they have backup power or interruptible contracts, without attributing the emergency condition to their demand.
• Routine economic demand response, price-induced voluntary reductions, general public conservation appeals, tests, and drills. Contracted demand response counts only when actually activated to address an actual or impending reliability emergency.
• Uncommanded datacentre trips, outages caused solely by local equipment failures, and curtailment of generation or electricity exports rather than customer consumption.
• Causal attribution made only by journalists, analysts, regulators, or other third parties rather than by the issuing operator.

Resolving source. The issuing grid operator's dated operational notices, emergency-procedure logs, official news releases, incident reports, or operator-authored regulatory filings; alternatively, a published report or order from the North American Electric Reliability Corporation (NERC) or the Federal Energy Regulatory Commission (FERC) that explicitly records both the instruction and the issuing operator's causal attribution. These are the permitted sources rather than the broader Public confirmation convention A2. Examples of operator publication channels include PJM Inside Lines and MISO Operations Displays. Apply the Resolution window and Resolution date convention A3. A retrospective confirmation may concern an earlier incident, but only qualifying publications within the Resolution window count.