The Southern Company: How Much Credit for Flexible Demand?
Download case PDF (5 pages)Request teacher’s pack
Information cutoff: October 2, 2026, end of day Eastern Time
Decision scope: Georgia Power Company’s future resource planning
Issuer context: The Southern Company (NYSE: SO) is the parent; Georgia Power is the regulated operating subsidiary in this case.
Document version: 1.0, October 3, 2026
Research basis: Literature review, source register, and evidence extraction.
The question
Georgia Power serves a fast-growing state while planning resources years ahead of when they can operate. A newly announced, very large customer brings a second question: how much should a utility count a customer’s promise to reduce electricity use when it plans for peak demand?
For this exercise, assume you advise a hypothetical future Georgia Power resource-planning review. The assignment is not a documented committee mandate. The Public Service Commission (PSC) approved Georgia Power’s 2025 Integrated Resource Plan (IRP), and an all-source request for proposals for 2,000–6,000 MW with projected 2032–33 delivery is already underway. Neither is being reopened here. Your task is to recommend how later planning updates should treat the flexible-demand commitment in the OpenAI agreement, while keeping the existing approved plan as the baseline.
Decision question: Should Georgia Power give OpenAI’s flexible demand response (DR) no planning credit until it is demonstrated, use a staged scenario-based credit subject to evidence gates, or count more of the contract’s maximum flexibility now when evaluating later discretionary resource needs?
Recommend a policy and specify the evidence and triggers that would change it. Do not assume a particular number of megawatts is dependable. The public record does not disclose the contract’s dispatch schedule, minimum load, penalties, or operating results.
A plan built before the customer announcement
Georgia Power’s 2025 IRP was approved by the PSC on July 15, 2025. The plan responds to the utility’s forecast of approximately 8,500 MW of load growth over six years, includes reserve margins, and combines generation, transmission, nuclear uprates, renewable resources, battery energy storage, and demand-side programs. That 8,500 MW is a utility projection, not measured future demand. [P2]
The next large procurement is not hypothetical. Georgia Power initiated an all-source RFP on June 9, 2026 for 2,000–6,000 MW with projected delivery or commercial operation in 2032–33. The June filing of final approved RFP documents and Georgia Power’s Q2 filing confirm that work is underway under the existing plan. [P1; P7]
The timetable matters. Generation and transmission resources can take years to plan, permit, build, and connect. Waiting may produce better information about demand and customer performance, but it can also leave less time to respond if forecasts rise or a resource is delayed. The question is how to handle future planning choices around that uncertainty, not whether the utility should ignore reliability requirements or cancel resources already approved or in procurement.
A large new load with a flexible portion
On July 22, Georgia Power announced a 25-year agreement to serve OpenAI’s planned Effingham County project. The utility described approximately 3,200 MW of new demand and said the agreement includes up to 1,000 MW of flexible DR during high-demand periods. Georgia Power said it may reduce energy delivered to the project “during certain times,” and suggested that flexibility could mean it invests in fewer new generation resources for future growth. The company also said OpenAI would pay the full infrastructure and electric-service costs and satisfy long-term energy-contract and financial-assurance requirements. [P3]
“Up to 1,000 MW” is the maximum disclosed flexible portion. It does not establish that 1,000 MW will be available at every peak hour, how long or how often Georgia Power may call on it, or whether the response has been tested. Nor does the public July announcement show that the utility can treat it like a dispatchable generator. The contract may be valuable; the planning value depends on when and how much load can actually be reduced, and under what operating conditions.
The company later described the contract as approved as part of the regulatory process. The underlying August 26 document is a joint statement by Georgia PSC Staff and Georgia Power. It says Staff reviewed the contract and confirmed it complied with the applicable large-load orders; Staff “does not object.” It is not a vote or order by the full Commission. [P4; P5]
The joint statement also sets out customer and ratepayer protections. Georgia Power must track large-load customer revenues, minimum bills, and forecast incremental costs. If the customer terminates early, the company agreed not to seek specified remaining-term incremental-cost revenue deficiencies from non-large-load customers in a later ratemaking proceeding. The Commission retains discretion over whether and how an unmitigated deficiency may be recovered, including from other large-load customers. Those terms address cost recovery; they do not resolve how much physical capacity the DR can reliably provide. [P5; P6]
The PSC filing record identifies a September 8 public contract-summary ZIP. That attachment was not readable in the research environment, so its detailed terms are not used here. Do not infer dispatch, compensation, or penalties from the public announcement.
The planning problem
Demand forecasts, resource needs, and customer flexibility are measured on different timelines. Georgia Power has an approved plan and an active procurement, but it will continue to assess needs as loads and resources develop. The OpenAI project could ramp on a different schedule than forecast. Its flexible portion could prove highly dependable, partially available, or constrained during the same hours when the system is tight. The public record does not quantify these possibilities.
Treating the full contract maximum as firm capacity now could support a smaller or later future build. If actual response falls short during a peak, the utility could face reliability pressure with fewer options to correct it. On the other hand, refusing any credit until a long operating history exists may cause the utility to commit too early to expensive and difficult-to-reverse resources. Even if the load is flexible, resource lead times could require decisions before a complete performance record accumulates.
The August regulatory statement introduces a separate financial question. Georgia Power’s press release reported approximately $950 million in annual customer savings beginning in 2029 from projected incremental revenue from its large-load customer portfolio and at least $15 per month in projected savings for a typical residential customer during 2029–31. These are company forecasts, not OpenAI-only amounts or realized savings. They do not establish the capacity value of OpenAI’s DR or a revised resource need. [P4; P5; P6]
Alternatives for a future planning update
| Policy | What the hypothetical review would do | Main benefit | Main risk or test |
|---|---|---|---|
| 1. No capacity credit until demonstrated | Keep the approved plan and active procurement as the planning baseline; count the agreement as new load, but do not offset forecast resource needs with its DR until operating evidence is available. | Protects against relying on an untested maximum commitment during a system peak. | Could understate a useful resource and encourage earlier or larger future commitments than necessary. How much lead time is needed to wait for evidence? |
| 2. Staged, scenario-based credit | Preserve the baseline, model low/partial/high availability cases, and allow only a later planning adjustment after agreed operational evidence and reliability review. | Makes uncertainty visible and creates a path to use proven response without assuming the maximum today. | Requires criteria for testing, verification, peak coincidence, and a fallback if performance or load ramp differs from plan. |
| 3. Earlier conditional credit | Use a larger share of the contract’s maximum in future need evaluations before a long operating record exists, conditional on the contract and regulatory reporting framework. | Could reduce or defer some future capacity commitments while the customer is expected to ramp. | The release does not disclose enough dispatch or remedy terms to establish dependable capacity. What evidence would make the risk acceptable before performance is observed? |
These are classroom policy choices, not options disclosed as formally under consideration by Georgia Power. Each must preserve approved reliability standards and all existing commitments. None changes the active 2032–33 RFP by itself.
Exhibit 1. Resource-planning baseline at the cutoff
| Item | Publicly disclosed position | What it means for this decision |
|---|---|---|
| 2025 IRP | PSC-approved July 15, 2025; projected approximately 8,500 MW load growth over six years; generation, transmission, reserve margins, renewables, BESS and demand-side programs included | Existing approved portfolio is the baseline, not a blank slate |
| 2032–33 RFP | Georgia Power initiated a 2,000–6,000 MW all-source procurement on June 9, 2026; approved documents filed June 11 | Procurement is underway; issuing it is not the open decision |
| OpenAI project | Approximately 3,200 MW of new demand; 25-year service agreement | Large load added to the planning outlook; ramp and realized demand are not established in the cited release |
| OpenAI flexibility | Up to 1,000 MW DR during high-demand periods | Contract maximum; dependable MW by hour/season and operating results are not disclosed |
Sources: 2025 IRP release [P2], Georgia Power Q2 Form 10-Q [P1], PSC RFP record [P7], and Georgia Power OpenAI release [P3]. Load-growth values are utility projections. The 1,000 MW is not converted into firm generation capacity.
Exhibit 2. What is known about the OpenAI agreement—and what is not
| Topic | Disclosed by Georgia Power or PSC Staff | Not established in the accessible public record |
|---|---|---|
| Scale and term | Approximately 3,200 MW of new demand; 25-year agreement | Actual ramp schedule, realized demand, and project operating profile |
| Flexibility | Up to 1,000 MW of DR during high-demand periods; Georgia Power may reduce delivered energy “during certain times” | Hours, season, peak coincidence, frequency, duration, notice, load floor, rebound, and testing method |
| Payment and assurance | Georgia Power says customer pays full infrastructure and electric-service costs and meets financial-assurance and long-term-contract requirements | Exact cost allocation, security amounts, dispatch compensation, shortfall remedies, and project-level net system costs |
| Regulatory process | Staff reviewed the contract, confirmed compliance with applicable orders, and did not object; joint statement provides for tracking/reporting and termination-cost protections | No full-Commission vote/order in the joint statement; no independently validated DR performance or avoided-resource value |
Sources: Georgia Power July 22 release [P3]; August 26 company release [P4]; PSC Staff/Georgia Power joint statement [P5]; August 27 incremental-revenue filing [P6]. The full Commission retains discretion in the treatment of an unmitigated revenue deficiency; the non-large-load customer protection applies to the specified early-termination cost deficiency described in [P5].
Exhibit 3. Separate customer-backed capacity context
On September 21, Georgia Power announced a separate subscription agreement with Google associated with approximately 96 MW of nuclear uprates and projected customer benefits over the units’ lives. The announcement said the arrangement remained subject to PSC approval. This is a different proposed customer-backed capacity mechanism; it is not equivalent to OpenAI’s DR commitment and is not a realized benefit. [P8]
Your assignment
Prepare a concise recommendation for the hypothetical future planning review. State which policy you recommend and why. Address:
- Which evidence should be required before Georgia Power gives DR capacity credit, and what reliability measure should it satisfy?
- How should the utility model the commitment while ramp, availability, and performance remain uncertain?
- What are the costs of waiting for an operating record, given resource lead times? What is the cost of assigning too much value to an unverified maximum?
- Which measurable triggers should increase, reduce, or suspend future planning credit?
Separate disclosed facts from your assumptions. Do not treat company-wide projected customer savings as OpenAI-specific savings, assume the full 1,000 MW is firm, or infer detailed contract terms from the unread public ZIP. The cutoff is October 2, 2026; Georgia Power’s Q3 results were scheduled for November 5 and are outside the case.
Source key
- P1: The Southern Company and operating-company Q2 2026 combined Form 10-Q, filed July 30, 2026; Georgia Power 2025 IRP/RFP discussion, printed p.46.
- P2: Georgia Power July 15, 2025 2025 IRP approval release.
- P3: Georgia Power July 22, 2026 OpenAI service announcement.
- P4: Georgia Power August 26, 2026 OpenAI contract/regulatory-process and large-load portfolio-savings release.
- P5: Georgia PSC Staff and Georgia Power joint statement, Docket 44280, document 228126, filed August 26, 2026.
- P6: Georgia Power August 27, 2026 large-load incremental-revenue filing.
- P7: Georgia PSC 2032–33 All-Source RFP final approved documents, filing 226846.
- P8: Georgia Power September 21, 2026 Google subscription announcement; subject to PSC approval.
Direct links and exact locators are in the research source register.