The First Data Center Utility Contract Disclosure Is Due

Georgia Power owes the Georgia Public Service Commission a public summary of the service agreement behind a 3,200 MW data center, and it owes it within ten days of a stipulation filed on August 26. That makes this the first data center utility contract disclosure we are aware of on a hyperscaler service agreement anywhere in the country. Every other large-load proceeding in this cycle has produced a tariff. This one produces a document.

The distinction is the whole story. A tariff tells you the pricing regime a class of customer will face. A contract summary tells you what one named counterparty actually signed, on what schedule, with what obligations attached. For anyone sizing equipment against announced load, those are different classes of evidence and only one of them is checkable.

What Georgia Power agreed to, and what it did not

The commitments come from a joint stipulation signed by PSC staff and Georgia Power attorneys. Per The Current’s reporting on the letter, Georgia Power “agreed to provide a public summary within 10 days” and will file “a semiannual public report that provides an overview of the performance of the company’s portfolio of large load data center customers.” The utility also agreed not to recover lost revenues from non-large-load customers if data center clients depart.

Read those two disclosure instruments carefully, because they are not the same thing and the difference has already been reported loosely.

The ten-day summary is a document about one contract. The semiannual report is an aggregate about a portfolio. A summary per counterparty tells you what a specific 3,200 MW load committed to. A portfolio overview tells you how a group of customers is performing in total, which is useful and much weaker. Georgia has not established a standing per-contract publication regime for future data center agreements, and anyone citing it as precedent for one is citing something that does not exist yet.

Worth noting on the same point: Georgia Power’s own press release announcing the approval carries the megawatts, the demand-response commitment and the savings figures, and says nothing at all about a public summary. The disclosure condition lives in the stipulation, not in the announcement. If you are tracking this, the PSC filing is the document to watch and the press release is not a substitute for it.

Why a data center utility contract disclosure beats a megawatt number

The unresolved question sitting under every load forecast is whether announced demand is contracted or aspirational. We have written before about how far apart those two words can be, with utilities in a single quarter publishing pipeline-to-contracted conversion ratios ranging from roughly 1.2:1 to about 9:1. A municipal utility or cooperative told to plan around a large-load number has had no instrument for checking which tier it belongs to.

A published service agreement summary is that instrument, at least for one project. Georgia Power’s release states the contract “includes 3,200 megawatts (MW) of new demand” and that OpenAI “has committed up to 1,000 MW of flexible demand response.” Those are the kinds of terms that either appear in a signed agreement with dates attached or they do not. A press release megawatt figure carries no such test.

The load is phased across 2028 to 2032 at a campus in Effingham County, near Savannah. For an equipment buyer, phasing is the number that matters more than the headline. A 3,200 MW load arriving over five years is a procurement calendar. The same figure quoted flat is a headline.

The cost-assignment clause is real, and it is not the news

Georgia Power’s release contains the sentence most people are quoting: “OpenAI will pay the full cost of the infrastructure needed to serve it.” That is a direct-assignment outcome, and it is the answer boards keep asking for when a large load lands in a service territory.

It is not new, and treating it as new gets the mechanism wrong. The Georgia PSC adopted a rule in April 2025 requiring data center contracts above 100 MW to be filed with the Commission at least 30 days before execution, and the developer-pays posture was already built into how Georgia handles these agreements. This approval confirms an existing rule working as designed. It did not create the rule. Several states are converging on similar answers through different instruments, including Pennsylvania’s “but for” large load tariff and the FERC show-cause order that put six RTOs on a compliance clock.

What is genuinely unresolved is what “the infrastructure needed to serve it” covers. Direct interconnection only, or upstream network upgrades as well? Where does the boundary sit, how is it measured, and what happens on early termination? A press release does not answer any of that. A contract summary might.

Three things to read the summary for

When the document lands, these are the terms with procurement consequences. A data center utility contract disclosure is only worth reading for what it settles, and it settles three things a press release never will.

One. The scope boundary on cost assignment. Whether the customer funds only the point of interconnection or also the upstream reinforcement determines whether an existing rate base absorbs the network work. That single boundary is the difference between a large load that pays its way and one that pays for its driveway.

Two. The delivery schedule and what triggers it. Phased load between 2028 and 2032 implies staged energization. If the summary carries milestone dates tied to contractual obligations rather than to intent, then the equipment demand behind it is on a clock, and that clock is already inside the lead-time window for step-up transformers and high-voltage breakers.

Three. What happens if the load does not show. Georgia Power agreed it will not recover lost revenues from other customer classes if these customers leave. The summary is where you would see whether that protection is a take-or-pay obligation, a termination charge, or a commitment with a shorter tail than the 25-year term implies.

One caution on how this large-load approval was produced

These conditions came out of a staff stipulation filed against a deadline, roughly an hour before a four o’clock cutoff for staff to approve or object. That is a negotiated approval, not a litigated record. Conditions bargained in the last hour tend to be the ones the utility could live with.

That is not a reason to discount them. It is a reason to be precise about the verb. Georgia did not impose a disclosure regime on Georgia Power. Georgia Power agreed to one, in exchange for an approval it wanted. The distinction will matter the first time advocates in another state cite Georgia as precedent and a utility there points out that the precedent is a voluntary commitment attached to a specific deal.

For buyers, the practical takeaway is smaller than the headline and more useful. One hyperscaler service agreement is about to become partially public. Read it for the scope boundary, the schedule and the exit terms, and treat what you learn as a template for the questions to ask the next time a large load appears in your territory. A single data center utility contract disclosure does not fix a forecast, but it does give you one verified point to calibrate the rest against. Then watch whether any other state converts this into a standing requirement, because that is the change that would make load forecasts checkable rather than merely published.

Large-load filings move faster than the coverage of them. The Feeder is our free monthly briefing on what actually changed in the record and what it means for equipment timing. Sign up here.

Related Reading

• Data Center Load Pipeline vs Contracted: How to Discount It
• Pennsylvania Large Load Tariff: ‘But For’ Cost Allocation
• FERC Large Load Interconnection Order: 6 RTOs on the Clock