Georgia Power gets approval for OpenAI’s 3.2 GW load as large customers support $950m annual savings

Southern Company (NYSE: SO) subsidiary Georgia Power has secured regulatory approval for its contract to serve OpenAI’s planned 3,200 MW project in Effingham County, transforming one of Georgia’s largest prospective electricity loads into a test of whether AI data centers can expand a regulated utility system without shifting their infrastructure costs onto households and small businesses. Under the 25-year agreement, OpenAI is required to pay the full infrastructure and electric-service costs associated with serving the facility and has committed to make as much as 1,000 MW of its load flexible during periods of high system demand. Georgia Power says the combination of the OpenAI agreement and other large-load contracts has increased projected customer savings from about US$556 million annually under an earlier framework to approximately US$950 million per year beginning in 2029.

The revised savings projection would produce approximately US$2.847 billion of total customer benefits across 2029 through 2031, according to Georgia Power. A typical residential customer using 1,000 kilowatt-hours per month is now expected to save at least US$15 per month, or approximately US$180 annually, beginning in 2029, compared with the US$102 annual benefit committed under the earlier regulatory settlement. That represents an increase of roughly 76% in the projected household benefit, although the savings remain forward-looking and depend on future large-load revenues, regulatory treatment and actual system costs developing as Georgia Power expects.

The approval arrives as utilities across the United States face an increasingly difficult question around artificial-intelligence infrastructure: whether tens of billions of dollars of new generation and transmission investment required for hyperscale computing will ultimately be paid by technology companies or shared with existing consumers. Georgia Power has taken the position that new large-load customers should cover the incremental cost of serving them while also contributing additional revenue toward the fixed costs of the broader system. The OpenAI agreement is unusually important because its 3.2 GW demand is large enough to make that policy economically visible rather than theoretical.

How large is OpenAI’s 3.2 GW Georgia electricity requirement?

A 3,200 MW requirement places the Effingham County project in a completely different category from a conventional commercial customer. For perspective, the OpenAI load equals roughly 32% of the approximately 9,900 MW of new generation and storage resources Georgia Power received approval to procure under its broader growth plan for 2029 through 2031. The comparison does not mean all 3.2 GW must be served by newly constructed generation dedicated solely to OpenAI, because utility systems combine existing generation, transmission capacity, purchases, storage and demand response, but it illustrates the enormous scale of a single AI campus relative to Georgia Power’s resource-planning programme.

OpenAI’s commitment to provide up to 1,000 MW of flexible demand is correspondingly significant. That flexible portion represents approximately 31.25% of the project’s maximum stated load, giving Georgia Power the ability to reduce electricity supplied to the facility during defined periods of extreme system demand. Instead of building enough generation to serve every megawatt of OpenAI demand during every system peak, Georgia Power can potentially rely on the customer to curtail part of its consumption when capacity is most constrained.

The flexibility provision changes the economics of the contract because generation assets are often built around the relatively small number of hours when system demand is highest. A customer capable of reducing up to 1 GW during those periods can lower the amount of incremental capacity the utility needs to construct or contract solely for reliability purposes. Georgia Power has consequently described the OpenAI arrangement as one of the largest single-facility demand-response commitments in the United States.

Why is Georgia Power projecting $950m of annual customer savings from large-load growth?

Regulated electric utilities recover much of their infrastructure cost from customers over long periods through approved rates. When a new data center requires substations, transmission upgrades, generation or other assets, the central affordability risk is that the project could fail, reduce demand or leave before enough revenue has been collected to recover those investments. Georgia Power’s large-load framework is designed to prevent that outcome by requiring long contract terms, minimum bills, financial assurances and customer responsibility for incremental infrastructure costs.

Southern Company’s published large-load framework calls for data center contracts of at least 15 years, minimum bills covering at least 100% of annual incremental costs to serve, termination payments connected to remaining incremental costs and collateral requirements linked to customer creditworthiness. OpenAI’s specific agreement runs for 25 years and requires the company to pay the full infrastructure and electric-service costs needed for its facility. These protections are intended to ensure that existing Georgia Power households and small businesses are not left financing stranded data-center infrastructure if projected demand changes.

Once those incremental costs are covered, revenue from large customers can help spread the utility’s broader fixed costs across a larger volume of electricity sales. Georgia Power previously committed that incremental large-load revenue would produce benefits of at least US$556 million annually after its next base-rate case, equivalent to approximately US$102 per year for a typical residential customer. The latest portfolio of customer contracts, including the approved OpenAI project, has increased the utility’s own projection to approximately US$950 million annually, a rise of roughly US$394 million or 71% from the earlier minimum benefit level.

Does the OpenAI contract remove the need for new Georgia generation?

No, because Georgia’s electricity demand is increasing far beyond one customer and the state is already pursuing a large expansion of generation, battery storage and transmission infrastructure. Georgia Power received approval in December 2025 to procure approximately 9,900 MW of new resources, including new combined-cycle gas generation, battery storage and other capacity intended primarily to support the 2029 through 2031 period. The utility subsequently launched planning for another 2,000 MW to 6,000 MW of dispatchable capacity for 2032 and 2033 as committed large-load customers continued to accumulate.

By April 2026, Georgia Power said 32 large-load customers had committed to approximately 15,600 MW of electric service, with 21 projects already under construction. OpenAI’s 3.2 GW project is therefore part of a much broader industrial and data-center expansion rather than an isolated demand spike. The scale explains why Georgia Power is simultaneously emphasizing affordability protections and pursuing one of the most aggressive utility resource-expansion programmes in the United States.

The flexible-demand provision helps reduce the marginal amount of generation needed for OpenAI, but it does not make a 3.2 GW project electrically insignificant. Georgia Power still must ensure sufficient transmission, local interconnection capability, energy supply and reliability across the project’s operating life, while OpenAI’s 1 GW flexibility is available only under the contractual conditions governing demand-response events. The economic benefit consequently depends on Georgia Power coordinating new generation and transmission investment with the timing of customer load rather than simply building every projected resource in advance.

How does the approval fit Southern Company’s broader earnings strategy?

Southern Company reported second-quarter 2026 earnings of US$1.2 billion, up from US$0.9 billion a year earlier, while adjusted earnings increased to US$1.3 billion. Management identified investment in regulated utilities, customer growth and usage among the drivers of higher adjusted earnings, showing that the data-center buildout is already influencing the financial narrative of the parent company even before many of the largest contracted loads are fully operating.

The opportunity is substantial because regulated utility investment can increase Southern Company’s rate base and future earnings when projects receive regulatory approval. The risk is equally clear: new generation and transmission require enormous capital commitments, and mismatches between projected demand and actual data-center construction could leave utilities seeking recovery for infrastructure that is not fully utilized. Southern Company specifically identifies uncertainties around significant electricity-demand growth driven by data centers and large-load customers, along with the requirement for substantial generation and transmission investment, as factors that can create capital-access and revenue-recovery risks.

Georgia Power’s approach attempts to reduce that risk by tying infrastructure costs and long-duration payment obligations directly to the large customers creating the demand. If those protections work as intended, Southern Company can expand its regulated investment base while simultaneously allowing part of the incremental customer revenue to flow back through lower rates. The OpenAI contract therefore matters not simply because it adds 3.2 GW of load, but because it provides a high-profile test of whether a utility can make AI infrastructure financially beneficial to both shareholders and existing customers.

Why is the $180 household saving still a projection rather than a guaranteed outcome?

Georgia Power’s US$950 million annual savings estimate depends on projected incremental revenue from large-load customers and therefore remains subject to future operating and regulatory conditions. The company itself cautions that actual results can differ because of electricity-demand changes, customer project schedules, infrastructure costs, regulatory decisions, commodity prices, financing costs and broader economic conditions. A technology company signing a large electricity contract does not by itself guarantee that every megawatt arrives on the original schedule or that every utility cost remains consistent with forecasts.

The timing also matters because the US$180 annual residential benefit is expected to begin in 2029, after Georgia Power’s next base-rate proceeding. Customers are already receiving a separate rate decrease approved in May 2026 worth approximately US$50 annually for a typical residential household, while base rates have been frozen under earlier regulatory actions. The future large-load benefit therefore sits on top of a broader sequence of rate decisions rather than representing an immediate reduction generated solely by OpenAI.

That distinction does not weaken the importance of the approval, because the contractual structure is what makes the future benefit plausible. OpenAI has agreed to pay its own infrastructure and service costs while supplying a flexible demand resource equal to almost one-third of its potential load, giving Georgia Power two ways to reduce the risk that existing customers subsidize the project. Whether those protections ultimately produce the full US$950 million annual portfolio benefit will become measurable only as the data-center buildout moves from signed contracts into actual electricity consumption.

What makes the OpenAI agreement a broader test for the U.S. power sector?

AI data centers are creating a political and regulatory problem because the infrastructure investment they require is arriving much faster than conventional utility planning cycles. Regulators in several markets are asking whether hyperscalers should pay special tariffs, commit to minimum demand or provide financial collateral so households are not exposed to stranded investments if projects are cancelled. Georgia moved early by tightening rules for very large energy users and structuring contracts around long commitments and incremental-cost recovery.

The OpenAI project gives that model unusually large scale. A 3.2 GW facility with 1 GW of flexible load is large enough to influence generation planning, transmission investment and future retail-rate calculations, while Georgia Power’s projected US$950 million annual savings give regulators and consumers a concrete number against which the strategy can eventually be judged. If the project proceeds and household savings materialize, Georgia could become evidence that data-center demand can reduce average system costs under sufficiently strong contractual protections.

If the underlying projects are delayed or infrastructure requirements prove more expensive than expected, the same case will illustrate the financial risk of planning utility systems around unprecedented growth forecasts. That unresolved tension is why the Georgia Power approval deserves attention beyond Southern Company and OpenAI: it is an early large-scale experiment in deciding who captures, and who pays for, the electricity economics of the AI infrastructure boom.