One contract, wearing a national program's name
Read the launch post in two layers. The first layer is a program: future OpenAI campuses, plural, designed to reduce electricity use when a grid is strained, with each site's arrangement setting how much demand can be cut, how quickly and for how long. The second layer is the evidence offered for it, and the evidence is a single agreement announced in July, for a single campus that has not been built.
That agreement is substantial on its own terms. Georgia Power's release puts the Effingham County facility's expected need at approximately 3,200 MW. The agreement provides for up to 1,000 MW of flexible demand response under a 25-year arrangement announced by Georgia Power on July 22, 2026. The 1,000 MW maximum represents approximately 31% of the facility's expected 3,200 MW demand. That is the contractual flexibility ceiling expressed as a share of projected site demand, not a guarantee that 31% of consumption will be reduced whenever the utility calls on the arrangement. OpenAI pays the full infrastructure and electric-service costs of serving the project, under rules for large loads that Georgia's Public Service Commission first approved in January 2025. The utility describes the arrangement as one of the largest single-facility demand-response commitments in the United States. If delivered at the stated scale, the ability to reduce up to 1,000 MW of demand would represent a substantial resource for grid management. The campus is still being developed, however, so the contractual commitment should not be confused with demonstrated operating capacity.
But a program is a claim about repetition. As of the launch post, there is no second campus, no second utility and no second figure. OpenAI says the arrangements will differ by location and that it will share further commitments as they are finalized, which is another way of saying the terms that would make this a program do not exist yet in public. What launched on November 2 is a national commitment built around one existing, publicly documented implementation. The Georgia agreement provides the first concrete example; the arrangements that would demonstrate the program operating across multiple campuses remain to be negotiated and disclosed.
The savings number belongs to a portfolio, and starts in 2029
The figure most likely to travel is $950 million. OpenAI's post attributes it carefully, and the care matters: Georgia Power projects approximately $950 million in annual customer savings beginning in 2029, from its portfolio of large-load agreements and expected growth, including this project. Three qualifiers sit inside that sentence. It is a projection. It covers a portfolio of large customers and general growth, not the Effingham campus alone. And it begins in 2029, three years from now, in a state where the campus itself is still in development.
The household version contains a discrepancy worth flagging. OpenAI says Georgia Power estimates savings of at least $15 a month for a typical residential customer using 1,000 kilowatt-hours. Georgia Power's July release, however, describes annual savings of $102 for a typical residential customer beginning in 2029, equivalent to about $8.50 a month. The public statements do not explain the difference in these estimates. Neither figure represents savings already delivered: both are forward-looking projections, not measured household bill reductions.
What the documents do not say
Neither public document specifies how often Georgia Power can call a cut, how long a single cut can last, or what happens to the campus's workloads when one is called. The Georgia release says reductions happen during certain times to support grid stability. That may be enough for a grid operator working under a filed contract. It is not enough for a reader trying to judge how much protection households actually have, because a gigawatt available twice a year and a gigawatt available on any hot afternoon are different products sold under the same headline.
OpenAI deserves credit for one sentence in its own post: flexibility cannot replace investment in new energy supply and transmission. That is the correct limit, stated by the company itself. The program should be judged against it. The test is not whether Effingham's terms are good; on the public record, they are unusually concrete. The test is whether a second campus signs terms a reader can inspect, with numbers for how much, how fast and how long. Until then, Grid Protectors is one good contract and a name waiting for its second example.
Sources
- [1] OpenAI — Introducing the Grid Protectors Program, November 2, 2026 (program launch and scope; site-by-site arrangements specifying how much demand can be reduced, how quickly and for how long; arrangements differ by location with further commitments to be shared as finalized; on-site generation or storage, batteries and power controls where local needs and equipment allow; families and small businesses should not subsidize a campus's electricity needs; commitment to pay full energy and infrastructure costs; Effingham County campus still in development and intended as pilot; agreement with Georgia Power includes up to 1,000 MW of flexible demand response, equated to the average demand of roughly 700,000 households; Georgia Power projection of approximately $950 million in annual customer savings beginning in 2029 from its portfolio of large-load agreements and expected growth including Project Camellia, with an estimated benefit of at least $15 a month for a typical residential customer using 1,000 kWh; US data center capacity projected to exceed 100 GW by 2030, citing BloombergNEF; flexibility cannot replace investment in new supply and transmission; read first-hand)Read source
- [2] Georgia Power — Georgia Power to serve OpenAI project in Effingham County, July 22, 2026 (facility expected to need approximately 3,200 MW; up to 1,000 MW of flexible demand response under a 25-year agreement; utility may reduce energy delivered during certain times to support grid stability; among the largest single-facility demand response commitments in the United States; OpenAI pays full infrastructure and electric-service costs with long-term contracts and financial assurances; updated large-load rules first approved by the Georgia Public Service Commission in January 2025; plan to provide annual savings of $102 for the typical residential customer beginning in 2029; read first-hand)Read source