Flexible demand response is a contract term with a concrete meaning for a large power user: the utility can reduce how much electricity it delivers to that customer when the grid is stressed, and the customer agrees in advance how much load can come down, how fast and for how long.
For a hyperscale or AI compute campus, that usually means pausing or shifting interruptible workloads, not flipping homes off. Training runs, batch jobs and other deferrable compute can be throttled so residential and small-business load stays up during heat waves, cold snaps or other peaks. OpenAI’s Grid Protectors framing is that kind of pre-agreed flexibility with utilities, regulators and policymakers, not a claim that every watt at a campus is optional.
How a utility curtailment contract works
In a special large-load agreement, the customer typically pays for the infrastructure and service needed to connect and serve the site. Separately, the customer may commit a block of flexible megawatts. When the utility calls the program, it reduces energy delivered to the facility within the contract rules. The customer is expected to shed that load by slowing or stopping interruptible processes.
That is different from ordinary energy-efficiency tips. It is also different from building a new power plant. Georgia Power said the OpenAI demand-response commitment “will also allow the company to invest in fewer new generation resources to serve future growth.” OpenAI itself wrote that “Flexibility cannot replace investment in new energy supply and transmission.” Load flexibility can reduce how much peak capacity a utility must build or buy; it does not itself create new megawatts for year-round growth.
The Georgia numbers
Georgia Power said on July 22 that OpenAI is expected to need about 3,200 megawatts for its Effingham County project and “has agreed to provide up to 1,000 MW of flexible demand response” under a 25-year agreement. Under that deal, “Georgia Power will have the ability to reduce energy delivered to the OpenAI facility during certain times to support stability of the electric grid.”
OpenAI’s Nov. 2 Grid Protectors post matches the flexibility figure: the Georgia Power agreement “includes up to 1,000 megawatts of flexible demand response,” which OpenAI equated to the average electricity demand of roughly 700,000 households. OpenAI said that once the campus is operating, it can reduce the power it draws “during times of need.” It also said it has committed to pay the full infrastructure and electric-service costs of the project.
So the 3,200 MW figure is Georgia Power’s stated expected need for the facility. The up-to-1,000 MW figure is the flexible demand-response commitment inside the same 25-year agreement, not the campus’s entire load being optional.
PSC approval and ratepayer framing
On Aug. 26, Georgia Power said the OpenAI contract had been approved as part of the regulatory process with the Georgia Public Service Commission. The utility said the contract was filed with the PSC in July and includes 3,200 MW of new demand, full infrastructure cost recovery from OpenAI, and up to 1,000 MW of flexible demand response.
Georgia Power tied the deal to a broader large-load portfolio that it projects will deliver about $950 million in annual customer savings beginning in 2029, or at least $15 a month for a typical residential customer using 1,000 kilowatt-hours. OpenAI repeated those portfolio savings figures and called Effingham a pilot for Grid Protectors even while the campus is still in development.
Axios, which has disclosed a content licensing and technology agreement with OpenAI, reported that the Grid Protectors pledge covers U.S. campuses OpenAI directly develops, with arrangements varying by site. The Verge reported OpenAI’s promise to free up power during extreme weather or emergencies.
Why labs pitch flexibility now
Public opposition to data centers’ power and water use has become a political and permitting issue. Midterm results underscored that pressure in several states, as AI Tech Daily reported. Utilities and large-load customers are answering with contracts that shift more cost and some peak risk onto the campus. Related grid-management debates include Texas’s earlier pause on new data-center hookups pending an audit.
Demand response does not settle whether a community wants a campus. It does define a measurable way a campus can shed load when the grid asks.
Story so far
Sources
- OpenAI: Introducing the Grid Protectors Program (November 2, 2026)
- Georgia Power: Georgia Power to serve OpenAI project in Effingham County (July 22, 2026)
- Georgia Power: Georgia Power contract with OpenAI approved; portfolio savings (August 26, 2026)
- Axios: OpenAI pledges data center power cuts, by Ben Geman (November 2, 2026)
- The Verge: OpenAI pitches data center power plants as a way to protect nearby areas, by Emma Roth (November 2, 2026)
- AI Tech Daily: OpenAI launches Grid Protectors Program (November 3, 2026)
- AI Tech Daily: Data center opponents emerge as big midterm winners in Texas and Georgia, Reuters reports (November 4, 2026)
- AI Tech Daily: Texas pauses new data-center grid hookups pending an audit (September 1, 2026)