The Ballot Won't Cancel the Build — It Will Decide Who Pays: Georgia Power's $16.3 Billion Test Case Before Nov. 3

Reuters' swing-state energy guide published Oct. 31, 2026 records the first cycle in which household electricity prices could be treated as a voter-level issue rather than a utility-industry issue. Average residential bills across eight key swing states rose about 7% year over year, and between 15% and 58% since 2019. In Pennsylvania, where the November contest for governor sits, the average household now pays roughly $182 a month for electricity, up 12% on the year and 58% since 2019. Ohio's bills rose 9.4% year over year and 53% since 2019.

US electric bills are up 10% since 2024, far outpacing inflation, and both parties now vow to control electricity rates. In the guide, voters in the eight swing states are named among the states seeing the sharpest increases.

Here is the mechanism that connects a voter to a turbine order, and it is not the one most people assume. Political pressure on customer bills does not stop capital spending. It re-allocates who pays for it. As of October 2026, 25 states have approved at least one large-load tariff and eight more have tariffs pending, and the architecture is remarkably consistent: a demand threshold, a minimum-bill or take-or-pay provision, an exit fee, and a collateral requirement.

That is the entire answer to the question voters are asking. Instead of a data center paying a wholesale price and leaving the network upgrade with the residential class, the data center signs a regulated contract that pays for the connection it causes, whether or not it ever materialises. The utility keeps building. The bill moves.

The second-order effect is on the allowed return itself. S&P Global's survey of major energy rate cases, filed with the Utah Public Service Commission, shows the average authorized return on equity for electric utilities slipping to 9.71% in the first quarter of 2026 from 9.84% for full-year 2025, with the trailing twelve months at 9.82%. The report attributes the direction to customer affordability concerns.

Against a 30-year Treasury that averaged 4.80% over the trailing year, a 9.82% authorized return leaves a spread of roughly 500 basis points, against about 800 basis points in 2020. The cushion over the long bond has already been substantially spent, and regulators are now spending the rest of it on behalf of voters. Volume growth is what is left.

On Dec. 19, 2025, the Georgia Public Service Commission voted unanimously, 5-0, to approve a stipulated agreement letting Georgia Power produce 9,885 megawatts of new energy — a build costing roughly $16.3 billion, with most of it serving large new customers such as data centers. At approval, Georgia Power said the plan would deliver about $102 a year in savings for a typical residential customer. By August 2026, with the PSC's approval of an OpenAI contract in Effingham County, that estimate had risen to about $180 a year, part of a portfolio delivering $950 million in annual savings.

Two commission seats are on the Nov. 3 ballot. In District 3, Democratic incumbent Peter Hubbard faces Republican Fitz Johnson; in District 5, Democrat Shelia Edwards faces Republican Josh Tolbert. Politico reported on Oct. 31 that Democrats are using voter anger over power bills and data-center growth to try to break Republican control of utility boards — and that outside Georgia and Arizona, those boards are appointed by the very governors whose races are now in play.

Virginia ran the experiment earlier than Georgia, and the order is the clearest single document in this story. On July 31, 2026, the State Corporation Commission required Dominion Energy to assign certain transmission infrastructure costs directly to the new large-load facilities that made them necessary, including mandatory contribution in aid of construction on direct-connect substations and lines. Dominion had sought to recover about $1.5 billion through the Rider T-1 transmission charge.

The lesson from Ohio is that the original load forecast was inflated — AEP cut its large-load forecast by more than 50% — but the surviving book is contract-grade rather than speculative. That distinction is the entire investment case. A utility with a speculative pipeline must ask the residential class for permission every time it builds. A utility with a contracted, collateralised book asks the data center, and the commission's job becomes approving the contract terms rather than the bill.

AEP's own numbers reflect the difference. Trailing diluted EPS of $6.13 and a 10.5% trailing return on equity sit above the 9.82% average authorised electric return — the signature of a utility whose growth is being paid for by its large-load customers.

Underneath every bill increase in the PJM footprint is a market that has stopped working. The December 2025 auction for the 2027/2028 delivery year cleared 134,478.1 MW of capacity — just 371 MW more than the prior auction, of which only 774 MW was genuinely new generation and uprates — and it cleared 6,516.6 MW short of PJM's reliability requirement. The July 2026 auction for 2028/2029 came in at 138,318 MW, but at the FERC-approved cap of $325 per megawatt-day, and it still fell 6,831 MW short, leaving a reserve margin of 14.7% against a 20% target.

This is where the election matters least and the money is most durable. GE Vernova ended the second quarter of 2026 with a gas-turbine order backlog of 116 GW, up from 100 GW in the first quarter, inside a total backlog of $176 billion. In the first quarter of 2026 alone, data centres generated more electrification orders than in all of 2025 combined. Second-quarter revenue was $11.1 billion with diluted EPS of $2.47 — the build converting into deliveries, not promises.

The non-obvious conclusion is that political risk in this cycle is priced into equity multiples, not into order books. Goldman Sachs expects US data-centre power demand to more than double to 66 GW next year from 31 GW, and PJM's inability to clear its capacity auctions is the physical confirmation. Turbine slots are booked years ahead against utility and hyperscaler commitments; the November vote changes the allowed return on those utilities' rate base, not the schedule at the factory. The equipment layer captures the value of a build that no ballot can cancel, and it does so with the lowest political beta in the chain.

The hyperscalers have responded to the new regime by buying long-dated, contract-grade power from the few suppliers that can offer it. Constellation Energy has signed 20-year nuclear power purchase agreements with Amazon, Alphabet, Meta Platforms and Microsoft. Vistra announced nuclear agreements with Meta on Jan. 9, 2026 supporting three plants in PJM, and its Comanche Peak agreement covers 1,200 MW. Constellation's trailing revenue of $31.3 billion and diluted EPS of $10.29 through the second quarter of 2026 sit against Vistra's $19.2 billion and $5.93.

The loser profile in this cycle is the regulated utility with a thin contracted book and a residential base already angry. Exelon is the clearest case: a trailing return on equity of 8.9% against the 9.82% average authorised electric return, capital expenditure running at 35.6% of revenue, and net debt at 5.8 times EBITDA. Its Potomac Edison subsidiary has filed a PC72 large-load rate schedule in Maryland under the Next Generation Energy Act and Utility RELIEF Act — the fight that matters, because without it Exelon recovers data-centre infrastructure from the residential class, in the one state where that class is already absorbing about $18 a month of capacity-driven increase.

That is the asymmetry. Over one to three years, the utilities that contract their load get paid to build while the political cost is exported to the data centre. The utilities that cannot get contracts are left asking voters for permission, and Nov. 3 tells them how that question is being answered.

The correct reading of the 2026 midterms is not that energy affordability is now partisan, but that it is now structural. A voter who sees a $180 annual saving on a Georgia Power bill, or a $0.94 transmission charge in Virginia, has been handed a specific mechanism for keeping AI load off the residential meter. Regulators, not politicians, wrote those mechanisms; the election decides who administers them.

That produces a hierarchy of exposure. Utilities holding binding, collateralised data-center contracts can grow rate base without a rate fight, and their earnings quality improves as the authorised return compresses. Utilities with a thin contracted book must fund the same build from the class that votes, and their returns sit below the allowed number. The turbine makers are paid either way. And the hyperscalers pay more for power in every state that adopts these tariffs, because a take-or-pay obligation is not free — it is the price of permission.