# Demand Response Reducing or shifting electricity consumption in response to grid conditions or price, instead of building generation to meet demand as it happens to fall. Economically it is the cheapest resource on the system, because not consuming a megawatt costs less than producing one. Institutionally it has always been the hardest to procure, because it requires many parties to do something rather than one party to build something. ## Why it matters more than its share suggests Electricity systems are sized for their peak, and peaks are extraordinarily concentrated. A large fraction of total capacity cost is driven by a small number of hours per year. Shaving those hours avoids capacity that would otherwise be built and paid for across the whole year. This is the point Severin Borenstein makes about data centers specifically: a large load that can flex during roughly the 1% of hours when the system is strained imposes dramatically less cost than an identical load that cannot. **The flexibility, not the size, is what determines the burden.** Two 500 MW campuses can have entirely different effects on everyone else's rates depending on this one property. ## Why AI load is unusually hard — and unusually promising Data centers have historically presented as the worst possible demand-response candidate: flat, constant, and contractually committed to uptime. A colocation operator selling availability guarantees cannot casually curtail. AI training changes the picture in both directions. Training runs are more interruptible than transaction processing — a checkpointed job can pause — and operators are geographically flexible in ways older facilities were not. But training load is also *volatile* in a way the grid has not had to handle before: rapid swings as jobs start and stop, which stress equipment and complicate forecasting. Reporting in August 2026 described those swings physically damaging batteries, generators and cooling systems faster than expected. So AI load is simultaneously more curtailable in principle and more disruptive in practice than what came before, which is why 2026 regulation attacks it from both ends: pay for flexibility, and mandate curtailability where payment fails. ## The 2026 instruments **FERC.** The June 2026 show-cause orders directed RTOs to develop new transmission services for large flexible loads able to limit withdrawals during system stress, including interim non-firm service while upgrades are under construction, with control and protection equipment installed at the customer's expense. Flexibility became a service class rather than a favor. **[[PJM Interconnection]].** Interim Resource Adequacy Service inverts the traditional curtailment order. From **2027-06-01**, new large loads that have not brought their own capacity are curtailed **before** pre-emergency demand response reaches other customers. The exemption is "bring your own new capacity." **[[ERCOT]].** Texas Senate Bill 6, effective June 2025, gave ERCOT authority to direct curtailment or backup-generation deployment for loads of 75 MW and above during or ahead of emergency alerts, and required **remote disconnect capability** for large loads connecting after 2025-12-31. **Emergency authority.** On 2026-05-18 the Energy Secretary issued Order No. 202-26-23 under Section 202(c) of the [[Federal Power Act]], authorizing PJM to direct large loads at or above 50 MW that hold backup generation to switch to it within 15 minutes of an emergency signal, with hospitals, 911 centers, water treatment, air traffic control and defense facilities exempt. > [!important] Curtailment order is a ranking of whose electricity matters least > Every one of these instruments answers the same question — when there is not enough power, who goes without — and answers it in advance, in a tariff. Historically that ranking was implicit and residential customers were protected last-resort by default. The 2026 designs make it explicit and put uncommitted large loads first. > > This is defensible: a load that arrived without bringing supply has a weaker claim than one that has been paying into the system for decades. But it is worth seeing clearly that a technical service definition is doing the work of a social priority ordering, and that the exemption list — hospitals, 911, water, air traffic control, defense — is a statement about what a society considers non-negotiable, written by an energy regulator. ## The historical irony Demand response spent thirty years as the underfunded, unglamorous corner of energy policy, championed by efficiency advocates against utilities that earned nothing from it. What finally made load flexibility a first-class product was not environmental policy but a capacity shortage caused by AI. The mechanism that arrived is also harsher than anything the efficiency movement proposed: not an incentive to reduce, but a rule about who gets cut off. ## See Also - [[Capacity Market]] — what flexibility substitutes for - [[Large Load Tariff]] — where curtailability is now written into service terms - [[PJM Interconnection]] — the IRAS design - [[ERCOT]] — statutory curtailment authority ahead of the federal regions