OpenAI says the campus it is planning in Effingham County, Georgia, can cut up to 1,000 megawatts of demand when the grid is strained. Against the all-time peak of the Southern Company system that Georgia Power belongs to, 38,194 MW on 17 January 2024 according to Southern Company’s 2025 annual report, that is about 2.6%, by our arithmetic. Against a Duke University estimate of how much flexible new load the Southern Company grid could absorb, 6.4 gigawatts, it is about 16%.
OpenAI launched its Grid Protectors Program on 2 November, calling Georgia its “starting point”, and we reported the pledge that day. This piece measures the 1 GW against the grid it sits on, and checks the “700,000 households” comparison OpenAI gives for it.
Business Pill · Percentiles
An evergreen Business Engineer explainer on percentiles: why an average can hide the extremes. It teaches the general idea only and does not discuss OpenAI or Georgia Power.
The key insight: By our arithmetic, OpenAI’s up-to-1,000 MW of flexible demand is about 2.6% of the Southern Company system’s all-time peak but about 16% of Duke’s curtailment-enabled headroom estimate for the Southern Company grid.
Georgia Power’s July release says OpenAI is “expected to need approximately 3,200 megawatts (MW) of power” for the facility and “has agreed to provide up to 1,000 MW of flexible demand response where it will reduce its power use in times of high demand”. By our arithmetic, that is up to about 31% of the campus’s load.
Under the 25-year agreement, the release says, “Georgia Power will have the ability to reduce energy delivered to the OpenAI facility during certain times to support stability of the electric grid”. The utility calls it “among the largest single-facility demand response commitments in the country”.
Georgia Power’s August release on the contract’s approval says the flexibility lets it “invest in fewer new generation resources to serve future growth”. OpenAI’s post adds that the arrangements for other campuses “will differ by location”.
OpenAI’s up-to-1 GW of flexible demand at Project Camellia, drawn inside five yardsticks: the campus’s 3.2 GW contracted load (Georgia Power); Duke’s curtailment-enabled headroom for the Southern Company grid at 0.25% curtailment (6.4 GW, Figure 1 label) and for eight Southeast grids (19.0 GW, our sum of the labels); Georgia Power’s six-year load-growth forecast in its 2025 resource plan (about 8.2 GW); and the Southern Company system’s all-time peak (38.2 GW, 2025 10-K). Shares are our arithmetic. The yardsticks measure different things and are not additive.
What Was Posted on X
OpenAI has launched its nationwide Grid Protectors Program, aiming to let utilities reduce electricity demand from its future AI data campuses during grid emergencies. Its planned Georgia campus includes up to 1 GW of flexible demand, roughly the average power use of 700,000 homes.
Choblin (@choblin29) summarised OpenAI’s Grid Protectors launch on X on 2 November, repeating the company’s comparison of the Georgia campus’s up-to-1 GW of flexible demand with the average power use of 700,000 homes. This piece checks that comparison against EIA’s and Georgia Power’s household averages and sizes the 1 GW against the Southern Company grid.
How Big Is 1 GW on This Grid
Southern Company’s 10-K for 2025 says maximum demand on its traditional electric operating companies, Southern Power Company and SEGCO was 37,006,000 kilowatts on 22 January 2025, and the all-time maximum was 38,194,000 kilowatts on 17 January 2024. The traditional operating companies are Alabama Power, Georgia Power and Mississippi Power. The figures exclude demand served by capacity retained by MEAG Power, OPC and SEPA.
By our arithmetic, OpenAI’s 1,000 MW is about 2.6% of that all-time peak and about 2.7% of the 2025 peak. The whole 3,200 MW campus is about 8.4% of the all-time peak.
The same 10-K says that since 2023 the traditional electric operating companies have contracted with new data centers and other large load customers “covering approximately nine GWs of electric load”. The 10-K was filed in February 2026, before the OpenAIproject was announced in July 2026.
When Georgia Power filed its 2025 resource plan in January 2025, it projected “approximately 8,200 megawatts (MW) of electrical load growth” over six years. By our arithmetic, 1,000 MW is about 12% of that figure, and 3,200 MW is about 39%. The plan predates the OpenAIproject, and the releases we read do not say how the contract fits into the forecast.
Checking the 700,000 Homes
OpenAI’s post says the 1,000 MW of flexible demand is “equivalent to the average electricity demand of roughly 700,000 households”. The comparison depends on how much a typical home uses.
The US Energy Information Administration says the average US residential customer bought 10,791 kilowatt-hours in 2022. Spread over the 8,760 hours of a year, that is an average draw of about 1.23 kilowatts, so 1,000 MW would cover about 812,000 such homes, by our arithmetic.
Georgia Power’s own “typical residential customer using 1,000 kilowatt-hours per month” draws about 1.37 kilowatts on average, which gives about 730,000 homes. OpenAI’s 700,000 implies about 1.43 kilowatts per home, or about 1,040 kilowatt-hours a month. By both yardsticks, the arithmetic gives more homes than OpenAI’s 700,000.
All three are averages over a whole year. None of the sources gives a home’s draw in the peak hours when the pledge would be called on.
What Duke’s Headroom Study Measured
In February 2025, Duke University’s Nicholas Institute published Rethinking Load Growth, which estimated how much new load the 22 largest US balancing authorities could add without building capacity, if the new load could be curtailed when needed. It calls this “curtailment-enabled headroom”.
The study found that “76 GW of new load” could be integrated at an average annual curtailment rate of 0.25%, “98 GW of new load could be integrated at an average annual load curtailment rate of 0.5%, and 126 GW at a rate of 1.0%”. At 0.25%, curtailment events averaged 1.7 hours, and the study counted an average of 85 curtailment hours a year across the 22 grids.
For the Southern Company balancing authority, which includes Georgia Power, the study’s Figure 1 labels the headroom at 6.4 GW at 0.25% curtailment and 7.7 GW at 0.5%. By our arithmetic, OpenAI’s 1 GW is about 16% of the first figure. Summing the labels for the eight Southeast grids outside the regional markets gives 19.0 GW at 0.25%, of which 1 GW is about 5%.
The two figures measure different things. Duke’s headroom assumes the whole new load can be curtailed as needed, based on 2016 to 2024 load data; OpenAI’s contract makes up to 1,000 MW of a 3,200 MW campus flexible. The 3,200 MW campus on its own equals half of the 6.4 GW, by our arithmetic.
How Others Have Set Up Flexibility
Google. In an August 2025 post, Google announced demand response agreements with Indiana Michigan Power and the Tennessee Valley Authority, calling them “the first time we’re delivering data center demand response by targeting machine learning (ML) workloads”. The post gives no megawatt figure, and says flexibility “is still in the early stages and will only be available at certain locations”.
Texas. The enrolled text of Senate Bill 6 of 2025 tells the state commission to set a large-load threshold of 75 megawatts and has ERCOT “competitively procure demand reductions from large load customers with a demand of at least 75 megawatts”. In an energy emergency, ERCOT may require large loads with on-site backup generation to “either deploy the customer’s on-site backup generating facilities or curtail load”.
PJM. On 16 January 2026, PJM’s board proposed “Avenues for new large loads to bring their own new generation or enter into a connect and manage framework subject to earlier curtailment”. In Texas the terms come from a statute and in PJM from a board proposal; in Georgia, from a contract between OpenAI and its utility that Georgia Power says was approved through the Georgia Public Service Commission’s regulatory process.
As we read it, the business lever is the price of a grid connection: Google’s post says flexibility lets large loads be interconnected more quickly, Georgia Power says it lets the utility build less, and OpenAI has now put a size on what it offers.
The Structural Read
As we read it, the share depends on the yardstick: about 2.6% of the system’s record peak, about 12% of Georgia Power’s six-year growth forecast and about 16% of Duke’s headroom estimate.
As we read it, the open question is duration: Duke’s curtailment events averaged 1.7 hours at a 0.25% rate, and the documents we read give no hours for OpenAI’s contract.
OpenAI, Grid Protectors post, 2 November 2026
“Flexibility cannot replace investment in new energy supply and transmission.”
Three Implications
FOR UTILITIES Georgia Power says the flexibility lets it invest in fewer new generation resources to serve future growth; its releases give no figure for how much less it plans to build.
FOR OTHER AI CAMPUSESOpenAI says its arrangements will differ by location, and Google’s 2025 post on its Indiana Michigan Power and TVA agreements gave no megawatt figure, so Georgia’s 1,000 MW is the one sized commitment in the documents we read.
FOR GRID OPERATORS Texas’s SB 6 sets a 75-megawatt large-load threshold for demand reductions, and PJM’s board proposed a connect-and-manage path subject to earlier curtailment.
The framework states: “Companies that solve the power equation will dominate the AI era; those that don’t will watch their chips sit idle.”
As we read it, OpenAI’s Georgia contract is one way of working on that equation from the demand side: up to 1,000 MW of its 3,200 MW campus can be turned down when the grid is strained. The framework was published in December 2025 and does not discuss OpenAI’s program.
The documents we read do not say how often, for how many hours a year or at what notice Georgia Power can call on the 1,000 MW, or what OpenAI is paid or charged for it. The Southern Company peak covers Alabama, Georgia and Mississippi Power, not Georgia Power alone. Duke’s headroom is a planning estimate from historical load, published before the OpenAI contract, not a study of this campus. The campus is still in development. We did not contact OpenAI, Georgia Power, Southern Company or Duke.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
OpenAI’s up-to-1,000 MW of flexible demand in Georgia is about 2.6% of the Southern Company system’s all-time peak, about 12% of Georgia Power’s six-year growth forecast and about 16% of Duke’s headroom estimate for the Southern Company grid, all by our arithmetic. Its 700,000-household comparison sits below what EIA’s and Georgia Power’s own averages give.
A note on sourcing. Figures come from OpenAI’s Grid Protectors post, Georgia Power’s July and August 2026 releases and its January 2025 resource-plan release, Southern Company’s 2025 10-K, the Duke Nicholas Institute’s Rethinking Load Growth report (Figure 1 data labels), EIA’s household electricity FAQ, Google’s August 2025 post, the enrolled text of Texas SB 6 and PJM’s 16 January 2026 release, all read on 2 November 2026. Shares, sums and household conversions are our arithmetic. We did not contact the companies.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
