OpenAI launches Grid Protectors Program with up to 1 GW of power cuts in Georgia
OpenAI announced the Grid Protectors Program on Monday, November 2, 2026, a commitment that the US data center campuses it develops directly will reduce their electricity use when local grids come under strain from extreme weather or other emergencies. The company said the program will be negotiated utility by utility, with each agreement setting how much demand a campus can shed, how quickly it must respond and how long the reduction can be sustained. Some sites may also send power back to the grid from on-site generators or batteries, although OpenAI described this as a possibility rather than a requirement for every campus.
The first site in the program is OpenAI's planned campus in Effingham County, Georgia, about 45 minutes from Savannah. Georgia Power expects the facility to need roughly 3,200 megawatts, and under a 25-year contract cleared by Georgia regulators in late August, OpenAI agreed to provide up to 1,000 megawatts of flexible demand response. OpenAI compares that 1,000 MW to the average electricity demand of about 700,000 homes, and it said the new program extends the Georgia arrangement into a national approach for future campuses.
The company framed the program around who pays for the grid. "Families and small businesses should not be asked to subsidize a compute campus's electricity needs," OpenAI wrote, adding that its ambition is to make this kind of flexibility a core part of how it builds and operates future campuses. OpenAI has so far named no site other than Effingham County, and it said the terms for each additional campus will be announced as they are finalized.
··········
WHAT THE OPENAI GRID PROTECTORS PROGRAM COMMITS TO.
According to OpenAI's statement, the Grid Protectors Program covers US data center campuses that OpenAI develops itself. At each of those sites, OpenAI will work with the local utility to agree three parameters: the amount of load the campus can drop during a grid emergency, the speed at which it can drop it, and the duration for which the lower load can be held. These are the standard building blocks of a demand response contract, the same mechanism utilities already use with steel mills, chemical plants and large commercial buildings, applied here to an AI compute campus.
OpenAI also said it will pay the full cost of the transmission lines and electric service infrastructure that its campuses require, so that the expense of connecting a multi-gigawatt facility is not spread across other customers' bills. In Georgia this commitment is already part of the contract filed with the state regulator, and the company presented it as a general rule of the program.
The statement leaves several operational points open. OpenAI did not publish a minimum curtailment percentage that every campus must meet, did not say whether on-site generation will be fossil-fired, battery-based or a mix, and did not specify notice periods. John McCarrick, OpenAI's head of energy policy, said the initiative reflects a shift in how the industry thinks about reliability, improvements in the technology and feedback from communities. He noted that a year ago a large hyperscaler was still defending "five nines" availability for its facilities, and said he does not think anyone is doing that now.
··········
THE EFFINGHAM COUNTY CAMPUS AS THE FIRST GRID PROTECTORS SITE.
The Georgia project predates the program by more than three months. Georgia Power announced on July 22, 2026 that it would serve the OpenAI campus, that OpenAI would pay the full infrastructure and electric service costs, and that OpenAI had agreed to provide up to 1,000 MW of demand response over a 25-year agreement. Georgia Power described the commitment as among the largest single-facility demand response commitments in the country, and said the ability to reduce deliveries to the site at times of high demand would let it build fewer new generation resources.
On August 26, Georgia Power said the contract had been approved, as part of a portfolio of large-load agreements that it expects to bring about $950 million a year in customer savings beginning in 2029. The utility put total projected customer benefits at $2.847 billion for 2029 through 2031, and said savings for a typical residential customer using 1,000 kWh a month would rise from the $102 a year it promised in December 2025 to $180 a year, or at least $15 a month. Those figures are company projections covering the whole large-load portfolio, and Georgia Power's release notes that actual results could differ materially.
........
Effingham County item | Figure | Source of the figure |
Expected campus demand | About 3,200 MW | Georgia Power, July 22 |
Flexible demand response | Up to 1,000 MW | Georgia Power and OpenAI |
Contract length | 25 years | Georgia Power |
Infrastructure cost | Paid in full by OpenAI | Georgia Power and OpenAI |
Contract approval | Announced August 26, 2026 | Georgia Power |
Portfolio savings | About $950M a year from 2029 | Georgia Power projection |
Typical home bill effect | $180 a year from 2029 | Georgia Power projection |
........
The contract text filed with regulators has been redacted, and Georgia Power has committed to publishing summaries of its large-load contracts and semiannual reports on its data center customer portfolio. Consumer advocates in Georgia argued during the review that the projected savings rest on assumptions that could still change, which is why the actual curtailment terms behind the 1,000 MW figure remain one of the most closely watched parts of the deal.
··········
HOW 1 GW OF DEMAND RESPONSE COMPARES WITH THE EFFINGHAM CAMPUS LOAD.
Placing OpenAI's figures side by side gives a clearer picture of what the Georgia commitment means in practice. The calculations below use only the numbers published by OpenAI and Georgia Power; they describe the maximum flexibility on paper and say nothing about how often, or for how long, the utility will call on it.
........
Measure | Value | How it was calculated |
Share of campus load that can be shed | 31.25% | 1,000 MW ÷ 3,200 MW |
Load that stays on during a full call | 2,200 MW | 3,200 MW − 1,000 MW |
Average demand per home implied by OpenAI | About 1.43 kW | 1,000 MW ÷ 700,000 homes |
Monthly use implied per home | About 1,040 kWh | 1.43 kW × 730 hours |
Full campus load in home equivalents | About 2.24 million | 3,200 MW ÷ 1,000 MW × 700,000 |
Annual savings per typical home | $180 ($15 a month) | Georgia Power projection |
........
The implied consumption of about 1,040 kWh a month per home is consistent with the 1,000 kWh typical residential customer that Georgia Power uses for its bill estimates, which suggests OpenAI's 700,000-home comparison is built on the utility's own average. The same arithmetic shows the scale of the campus: at full load it would draw roughly as much power as 2.24 million average Georgia homes, and even with the full 1,000 MW curtailed it would still be consuming 2,200 MW.
··········
WHAT GRID PROTECTORS MEANS FOR UTILITIES, RATEPAYERS AND CHATGPT USERS.
For utilities, the program offers a template. A data center operator that agrees in advance to curtail a defined block of load lets the utility plan its peak capacity around a smaller number, which is the basis of Georgia Power's argument that it can invest in fewer new plants. OpenAI's promise to pay for its own grid connections addresses the second concern that utilities and regulators have raised about AI campuses, namely the risk that ratepayers end up carrying the cost of lines built for a single customer.
For households near future OpenAI campuses, the practical effect depends on contracts that do not exist yet. The Georgia numbers are the only ones on the table, and they are projections for 2029 onward. Residents in other states will learn what OpenAI has committed to only when the company and the relevant utility publish the terms for each new site, which OpenAI said it will do as agreements are finalized.
For people who use ChatGPT, Codex or the OpenAI API, the company has not explained how a curtailment event at one campus would affect service. One plausible approach, and it is an inference rather than anything OpenAI has described, would be to slow or relocate interruptible work such as batch jobs and training runs during a curtailment while keeping interactive services running elsewhere; OpenAI has not said which workloads it would reduce. Readers following the scale of OpenAI's infrastructure plans can compare this program with the company's earlier Stargate expansion with Oracle and its four new agreements in Abu Dhabi, which add capacity outside the United States.
··········
THE POLICY PRESSURE BEHIND OPENAI'S DATA CENTER POWER PLEDGE.
The announcement arrives as rising electricity bills have turned data center construction into a local political issue across the United States, with several projects delayed by opposition from residents and officials. The Federal Energy Regulatory Commission is working on rules for how very large loads connect to the transmission system, and a Senate permitting bill would allow states to require data centers to limit their power use as a condition of grid service. The White House has also asked developers to sign a voluntary pledge to cover their own power costs.
Other companies are moving in the same direction. Google and NVIDIA have formed their own alliance focused on making data center demand more flexible, and Georgia's Public Service Commission adopted dedicated rules for large-load customers in January 2025, the framework under which the OpenAI contract was reviewed. OpenAI's program puts its name on a set of commitments that regulators were already starting to demand, and gives it a public standard to point to when it seeks approval for new campuses.
OpenAI itself acknowledged the limits of the approach. The company said flexible demand cannot replace new power plants and new transmission lines, which the grid will still need as AI campuses grow. Demand response reduces the peak a utility has to plan for, while total energy consumption over a year remains largely set by how much computing the campus runs.
··········
OPENAI GRID PROTECTORS PROGRAM: CONFIRMED FACTS AND OPEN QUESTIONS.
What OpenAI has stated is clear in outline. The Grid Protectors Program applies to US campuses that OpenAI develops directly, it commits those campuses to reduce demand during grid emergencies under terms negotiated with each utility, and OpenAI says it will pay for the infrastructure its sites require. The Effingham County campus is the first site, with an approved 25-year contract, an expected load of about 3,200 MW and up to 1,000 MW of flexible demand response, while the $950 million a year in savings from 2029 is Georgia Power's projection for its entire large-load portfolio.
The open questions are mostly about enforcement and detail. OpenAI has not named a second site, has not published response times or maximum curtailment durations for Georgia, and has not said how often it expects to be called on. The Effingham campus is still being developed, so the 1,000 MW commitment is a contractual capability that has not yet been exercised. The program's value to ratepayers will be measured by the published contract summaries and by how the campus actually behaves during Georgia's first summer and winter peaks after it comes online.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
