The real AI referendum on today's ballot isn't about whether to build

Beneath the marquee contests on today's midterm ballots sits a quieter, stranger election: dozens of local votes and utility-commission races that amount to a referendum on who pays for the artificial intelligence boom's power bill. Voters in Janesville, Wisconsin, are deciding today whether to require a public vote before a former General Motors transmission-plant site can become a data center. Georgians are choosing two members of the Public Service Commission in a race explicitly fought over Georgia Power's six rate increases since 2023. A ban measure is on the ballot in Monterey Park, California. These contests rarely make a national headline, but together they are a more honest gauge of how the country feels about the AI build-out than any stock index.

The anger behind these votes is justified. The remedy most of them offer is not. A ban or a moratorium feels like accountability, but it does nothing to fix the actual mechanism that has been pushing data centers' costs onto ordinary ratepayers, and it leaves that mechanism in place for the next large customer that comes along. The harder, less satisfying fix, already being built in a handful of states and nudged along by federal regulators, is to rewrite the tariffs so big power users pay for the capacity they actually draw. That is the real test the country should be applying to this moment, and most of today's ballot measures do not pass it.

A boom that shows up on the bill before it shows up anywhere else

The scale of the demand is not in dispute. Federal energy forecasters opened this year by projecting that U.S. electricity use would climb for a fourth consecutive year, something that has not happened since 2007, with the fastest four-year growth since 2000. The driver, in the government's own words, is concentrated and specific.

Demand is "driven largely by increasing demand from large computing facilities, including data centers."

That single sentence, from the Energy Information Administration's own administrator, is why this argument cannot be waved away as NIMBYism or tech-skepticism. It is why regional grid operators are seeing capacity prices move the way they are: one independent market monitor's analysis, reported as a 76 percent year-over-year jump in PJM Interconnection's wholesale capacity prices, attributed most of that spike to data-center load rather than ordinary growth. And it is why, in Georgia, a commission that most voters could not have named five years ago is now the center of a competitive statewide race, after approving a 10-gigawatt expansion of Georgia Power's system through 2030 on top of a run of rate increases that Democrats are campaigning against directly, as reporting on the party's strategy there lays out.

Even federal regulators have effectively conceded that the pricing system was not built for this. In December, the Federal Energy Regulatory Commission found that PJM's tariff was "unjust and unreasonable" because it lacked clear rules for how data centers co-located with power plants should be billed for the grid services they still use. If the agency that oversees wholesale power markets concluded the rulebook did not fairly assign these costs, it is not unreasonable for a homeowner in suburban Atlanta or exurban Wisconsin to have reached the same conclusion looking at a monthly statement.

Why a ban is the wrong instrument

The instinct to reach for prohibition is understandable, and it has already produced real results. Port Washington, Wisconsin, voted roughly two-to-one in April to require a public referendum before the city can award tax breaks to large developments, a direct response to a $15 billion data-center project already under construction there; it arrived too late to touch that project, but it has reshaped how the next one gets approved, as coverage of that vote and today's follow-on measure in Janesville makes clear. Ballot trackers count roughly three dozen data-center-related measures nationwide this cycle, concentrated heavily in Ohio, where more than a dozen communities are voting on local approval requirements, according to a tally of this year's data-center ballot measures.

There are three problems with treating these votes as the fix rather than a symptom. First, a local ban or a one-time referendum does not reach the capacity that is already built or already approved; Georgia's 10-gigawatt expansion is locked in regardless of who wins the commission seats today. Second, a ban that succeeds in one town simply redirects the project to the next county or the next state, which does nothing to change the underlying rate design and may leave that jurisdiction with weaker rules than the one that just said no. Third, and most importantly, the people pushing hardest for bans are often responding to a legitimate complaint about capture and speed, not actually choosing prohibition over sound tariff design as a matter of principle. Critics of Georgia's commission have pointed to how heavily utility-linked contributions fund incumbents' campaigns, a fair complaint about who writes the rules. But the answer to a captured rulemaking process is a better rulemaking process, not a permanent local veto that leaves the statewide rate structure untouched for everyone not lucky enough to live in a town that voted first.

The model that is actually working

Oregon offers the clearest counter-example. Under a 2025 state law, the Oregon Public Utility Commission this year approved a new rate class for large data centers served by Portland General Electric, and the numbers show what cost-causation pricing looks like in practice: the commission's own order raises average rates for qualifying data centers by about 29 percent while cutting residential rates by roughly 1.3 percent, with penalties if a large customer draws more than its contracted capacity. Oregon's governor, in publicly endorsing the commission's decision, made the logic explicit: growth should pay for the growth it causes, rather than spreading those costs across everyone's bill. That is a regulatory body doing the unglamorous work of reallocating cost to its source, and it did not require banning a single data center to do it.

FERC's December order points the same direction for the mid-Atlantic grid, directing PJM to build new transmission-service categories that bill co-located large loads for what they actually withdraw rather than letting them ride on terms designed for ordinary customers. Neither fix is simple, and neither arrived quickly enough to spare this year's ratepayers. But both treat the complaint seriously as a pricing problem, which is the only way it actually gets solved. A short list of what distinguishes the two approaches is useful:

• A ban stops a specific project; a cost-causation tariff governs every future project in the territory, including the ones a ban cannot see coming.
• A ban is decided once, at the ballot box, with no mechanism to adjust as load patterns change; a tariff can be revised through the same regulatory process that set it.
• A ban leaves the existing rate structure, and its hidden subsidies, fully intact for every other large customer still connected to the grid.

What tonight's results will actually mean

Whoever wins the Georgia commission seats, and whatever happens in Janesville and the rest of Ohio's crowded ballot, the underlying arithmetic will not change by morning. Electricity demand tied to computing is still climbing toward the record levels forecasters described at the start of this year, and the states that have not yet separated that demand's cost from everyone else's bill will keep fighting this exact battle every election cycle until they do. Georgia's commissioners can talk about data centers paying their "fair share," but the only way to make that more than a slogan is the kind of explicit, published rate math Oregon's regulators just produced. Voters are not wrong to be furious about what has shown up on their bills. They would do better to direct that fury at the tariff filings their state commissions approve in obscurity than at the data centers themselves, because prohibition is the one tool on today's ballot that cannot touch the bill already in the mail.