OpenAI’s Grid Protectors program starts from a Georgia campus that can shed 1,000 megawatts
A post dated Monday says future compute campuses will cut demand when the grid is strained. The Effingham agreement, already described by Georgia Power in July and approved in August, includes up to 1,000 megawatts of flexible demand response. The $950 million savings figure is the utility’s portfolio number, beginning in 2029.
Feature4 minIn-depth
James Okonkwo4 min read
- OpenAI
- Energy
- Georgia
- Data centers

Photo: OpenAISource page
OpenAI on Monday published the Grid Protectors Program, a plan for the compute campuses it builds to use less electricity when the grid is under strain. The post, dated 2 November, says Georgia is the starting point. For the planned campus in Effingham County, OpenAI says it has committed to pay the full infrastructure and electric-service costs, and that the agreement with Georgia Power includes up to 1,000 megawatts of flexible demand response. The company says that figure is equivalent to the average electricity demand of roughly 700,000 households, more than Washington, D.C., and Boston combined, and that the practical use in Georgia is to leave power for air conditioning during extreme heat.
The campus is still in development. OpenAI says it wants the same kind of flexibility to be a core part of how later campuses are built and run: facilities designed to reduce use when needed, and agreements with utilities on how much demand can come off, how quickly, and for how long. The post says those arrangements will differ by place, and that further commitments will be shared as they are finalized. Depending on local needs and the equipment on site, it says the company will also work with grid operators to supply power from on-site generation or storage, and will use batteries and power controls to smooth changes in demand. It cites a projection that U.S. data center capacity will exceed 100 gigawatts by 2030.
Ben Geman at Axios, writing Monday, described the pledge as covering U.S. campuses OpenAI directly develops, and as applying when extreme weather or other emergencies strain the grid. His story said the blog post would appear later that day, and it quoted the lines about making flexibility a core part of how future campuses are built and operated. The post that is up uses that language. Geman also quotes John McCarrick, OpenAI’s head of energy policy, on a shift away from an older industry standard.
I was in a meeting about a year ago with one of the other large hyperscalers, and they ... were defending the five nines, and I don’t think anybody’s doing that now.
— John McCarrick, OpenAI’s head of energy policy, to Axios
Axios explains “five nines” as the idea that a data center must be available 99.999 percent of the time. The blog post does not use the phrase. It says families and small businesses should not subsidize a campus’s electricity, and that OpenAI has been committed to paying for the power it uses and for the infrastructure that serves its sites.
A contract Georgia Power described in the summer
The Georgia figures Monday’s post relies on were already public. On 22 July, Georgia Power said the project, which it placed about 45 minutes outside Savannah, was expected to need about 3,200 megawatts, and that under a 25-year agreement the utility could reduce the energy delivered to the facility at certain times. It called the load-flexibility agreement one of the largest single-facility demand response commitments in the country, and said the arrangement would let it invest in fewer new generation resources. Aaron Mitchell, senior vice president of strategic growth, said companies like OpenAI were “making real commitments when it comes to responsible operation and development.”
On 26 August, Georgia Power said the Georgia Public Service Commission had approved the contract. That release says the contract includes 3,200 megawatts of new demand and up to 1,000 megawatts of flexible demand response, and that OpenAI will pay the full cost of the infrastructure needed to serve it. The dollar figures in the same release are for a wider set of customers. With the projects announced to date and projected future growth, Georgia Power said a typical residential customer using 1,000 kilowatt-hours a month was expected to see savings of at least $15 a month beginning in 2029, which it described as $180 a year, up from a December 2025 commitment of $102 a year. It put the projected incremental revenue from new large-load customers at about $950 million a year, and the customer benefit at $2.847 billion over 2029, 2030 and 2031. Kim Greene, the company’s chairman, president and chief executive, said that in December the commitment had been $102 a year for a typical family beginning in 2029, and that the expected figure was now $180.
OpenAI’s Monday post points at that portfolio. It says Georgia Power projects about $950 million in annual customer savings beginning in 2029 from its large-load agreements and expected growth, including a project the post calls Project Camellia, and a benefit of at least $15 a month on the same typical bill. The post says the Effingham agreement contributes to the utility’s broader plan to lower costs. It also says flexibility cannot replace investment in new energy supply and transmission, and that long-term contracts and financial assurances can protect other customers if a project is delayed, reduced or canceled.
Axios adds two pieces of context that sit outside the blog post. It says a Federal Energy Regulatory Commission demand that grid operators justify or reform the procedures for connecting large power users includes provisions that promote flexible demand, and that a Senate permitting bill would let states require data centers to limit power use at times as a condition of receiving grid service. Those are Axios’s descriptions of proceedings and a bill. They are not terms printed in the Effingham releases.


