OpenAI’s announcement that it is building a massive data center campus near Savannah has put a familiar question back in the spotlight: do data centers pay their own way, or does their growth end up on everyone else’s electric bill?
The project, called Project Camellia, would sit on 1,400 acres in Effingham County and draw up to 3.2 gigawatts (GW) of power from Georgia Power, delivered in phases between 2028 and 2032. That means it would require as much power as over half the homes currently in Georgia, in less time than it takes to bring one new gas-fired power plant online.
OpenAI says it’s investing $20 billion in the site and has a contract with Georgia Power pending before the Public Service Commission (PSC) staff for review and approval or referral to the full PSC. The company and the utility are both quick to say that existing customers won’t foot the bill for this one.
So, is that true? Unfortunately, the honest answer is: we don’t know, and the odds aren’t great. Here’s what we do and don’t know, and why it matters.
How Data Centers Impact Your Bill
Utilities and state regulators typically divide the cost of power plants between three customer classes: industrial, commercial, and residential. Data centers are billed as industrial or commercial customers. You and I are billed as residential customers.
Georgia Power’s last rate case was in 2022. That case found that residential customers, based on usage patterns, should be allocated roughly half the cost of the company’s power plants, with the other half spread across all commercial and industrial customers. The total (for all of Georgia Power) peak-hour energy demand used in that study was about 16 GW.
Now, do the math on adding a single 3.2 GW data center. Adding that much new demand pushes total system peak from 16 GW to 19.2 GW. If residential peak usage stays flat, residential customers’ share of power plant costs would drop from about 50% to about 42%.
That may sound like residential costs will go down, but the “pie” of costs that has to be split up will also rise significantly: building new power plants now costs twice as much as older ones due to the market rush to supply data centers. Depending on how much data centers pay (which is a secret), residential customers footing around 42% of the new, higher costs may pay more than before, when they were charged for 51% of a smaller pool of costs.
There are numerous other variables that can affect these calculations, but there is one scenario that would be clearly bad for residential customers: what if the utility spends billions of dollars building new power plants and then the data center bubble bursts and they chip in very little? Then, the residential customers would face the worst of both worlds–paying a higher percentage of total costs for new higher-cost power plants.
Data Center Contract Terms Matter
Large loads like data centers sign service contracts with Georgia Power. These contracts typically run around 15 years and can include:
- Minimum bill provisions, which require the customer to pay for a baseline amount of electricity whether they use it or not; and
- Exit fees or penalties if the data center uses less power than expected, delays its buildout, or leaves the contract early.
These terms are particularly important in contracts for electricity service for data centers in the artificial intelligence (AI) sector because that sector is in flux, either growing fast or on the verge of a bubble, depending on who you ask. Since the data centers also use a lot more electricity than other large load customers, utilities have to build out a lot of long-lived infrastructure to serve those customers. If the data centers don’t show up or leave early, the utility has to recover those costs from the rest of its customers.
Georgia PSC Role
The Georgia PSC has taken steps in the right direction, but the current process leaves too little discretion to Georgia Power and lacks transparency.
The PSC approved new rules on January 23, 2025, that took effect February 1, 2025. The rules:
- Apply to large loads of 100 MW or greater. Any new customer at or above that threshold must sign a customized contract rather than use a standard tariff.
- Allow Georgia Power to use a minimum bill, but do not specify how that minimum bill is calculated.
- Allow longer contract terms, up to 15 years, but it was later revealed that if a contract is terminated early, only 2 years of minimum bills are required, not the full length of the remaining contract.
- Require Georgia Power to submit new large-load contracts to the PSC for review.
- Direct Georgia Power to demonstrate that data center revenue is reducing, not increasing, residential bills, and to file quarterly reports tracking new large loads.
Georgia Power has already made, and fallen short of, a version of this promise once before. In the 2023 Integrated Resource Plan Update (Docket No. 55378), the company agreed that revenue from new large-load customers would put at least $2.89 per month of downward pressure on the typical residential bill for 2026–2028, and committed to filing a cost-of-service study showing how those costs and revenues get allocated across rate classes. But according to testimony in a more recent capacity certification proceeding, Georgia Power has never produced verifiable evidence that it met the $2.89 commitment, telling regulators at one point that it doesn’t have the calculation. A newer, larger $8.50-per-month promise for 2029–2031 rests on the same kind of language, with no clearer mechanism for enforcement.
These are real improvements over what existed before. But better than nothing isn’t the same as sufficient. Here’s why:
- The rules assume the starting point was fair. When Georgia Power and the Commission say a data center is “paying its fair share,” that claim rests on the assumption that the underlying cost allocation was correct in the first place, unless and until regulators create a distinct asset class or allocation method specifically for data centers. If the 2022 rate case’s allocation was already imperfect, layering data center billing terms on top of it doesn’t fix that.
- In addition to using data center load to justify new generation and transmission investments, Georgia Power has also used it to justify delaying the retirement of old, dirty, and expensive coal plants. Add to that the fact that the new gas plants the PSC approved in late 2025 are especially expensive, and it’s difficult to see how the current situation meets the Commission’s intent for large loads to reduce residential bills.
- The math for allocating costs across sectors changes when one sector drives most of the growth. Historically, load growth was spread across many customers and classes, which is part of why cost allocation formulas assume gradual, broad-based change. When one sector, in this case AI data centers, accounts for a disproportionate share of new demand, the old formulas need to be revisited, not just applied with a few extra contract terms bolted on.
It is important to acknowledge the intent of the Georgia PSC staff in this whole process. We genuinely believe their intention is to reduce the risk of higher bills on existing customers. However, the PSC staff cannot have enough time and information to evaluate whether a data center is paying its fair share under a fully redacted contract in just 30 days and without a new cost of service study. And the current process completely leaves out other stakeholders.
Just How Big is 3.2 GW?
It’s easy for gigawatts to become an abstraction, so here’s some context specific to Georgia: the average Georgia residential customer uses about 1,035 kWh per month, or roughly 1.4 kW on average. At 3.2 GW running continuously, Project Camellia alone is projected to eventually draw as much power as roughly 2.3 million Georgia households — more than half of the state’s approximately 4.1 million households, and more than exist in the entire Atlanta metro area.
That’s not a reason to reflexively oppose the project. But a load of this size, added to a system that’s already growing quickly for other reasons, is exactly the kind of case where we need a clear and transparent process that takes the time it needs to verify the benefits to all customers, and mitigates potential risks of even higher bills in the future.
Tell the PSC Staff This Needs More Review
Last November, Georgia voters sent a clear message to their Public Service Commissioners that the decisions they make matter when two Commissioner incumbents lost their re-election bids. The PSC’s attempt to protect existing customers is a step in the right direction. But with data centers as big as Project Camellia, 30 days is just not enough time for PSC staff to review a contract.
The PSC staff should refer the contract for review to the full Commission, and the Commission should take another look and implement rules that:
- Ensure, not just recommend, data centers pay their fair share;
- Require transparency and data-backed analysis, not just telling the public to trust Georgia Power; and
- Take the time needed to get this right.
The PSC staff has until August 26 to refer this contract to the full Commission. Make your voice heard and tell them this contract deserves further review. And know where candidates for Georgia PSC stand on this and other issues you care about before voting this November.
Yes, we hear every day that AI companies are in a race, but the stakes are too high to “move fast and break stuff” if the things being broken are family and business budgets because electricity becomes unaffordable.
