Data center bans swept the ballot — Texas flipped, 13 GW at risk
For two years the fight over AI infrastructure was fought in planning commissions and interconnection queues. On Tuesday it was fought at the ballot box, and the industry lost nearly every vote it was in.
Voters restricted data center projects in almost every local measure put before them, and Texas — the state expected to host a quarter of the American data center capacity coming online by 2030 — elected a Democratic governor who campaigned on halting new builds.
404 Media's Jason Koebler counted 40 data center measures on Tuesday's ballot and voters restricting 39 of them, citing the Data Centers on the Ballot tracker. Other tallies differ in the details rather than the direction: Jim Bianco counted 30 measures with restrictions leading 29 to 1, and the Washington Post put the result at more than 20 bans or restrictions passed. The counts diverge because trackers disagree on whether advisory questions and tax-break conditions are restrictions. The margins don't diverge.
Those margins are the story. Trenton, Ohio passed a charter amendment banning data centers over 25 MW by 80-20 — the town where AWS is building a 250 MW facility and had proposed a second campus on land the city tried to annex. Urbana, Ohio voted 73-27 for a ban, in a town where Thor Equities planned a $1 billion campus and is already suing the city over the moratorium it enacted in March. Perry, Ohio passed 87-13 a measure that forces hyperscalers seeking tax breaks to post bonds and meet noise limits. Defiance approved a 25 MW threshold ban. Hubbard was carrying its ban 76-24, with Crane Data Centers' proposed 300 MW project already declared dead by the mayor. Emporia, Kansas rejected a 1,000-acre, gigawatt-scale project 60-40, and in Madison County, Illinois, voters in four townships rejected advisory questions on allowing data centers by wide margins.
None of those are symbolic. Each one lands on a specific project with signed land contracts and, in several cases, an interconnection agreement already in a utility's queue. We previewed this map before the vote — data center moratoriums are on the ballot in 16 governor races — and the turnout made the local measures look less like a protest vote than a policy.
Then Texas, which matters more than the rest of the list combined. James Talarico beat Ken Paxton for the state's first Democratic Senate seat in 38 years, and Gina Hinojosa won the governor's mansion. Hinojosa campaigned on halting new data center development unless operators forgo subsidies, commit to union labor, drop nondisclosure agreements and show that household power and water bills will not rise. Elon Musk, who spent more than $47 million backing Paxton according to the Financial Times, ended the night posting that "data centers in space looking pretty appealing right now."
The arithmetic behind that joke is the number to remember. Gordon Johnson of GLJ Research put the Texas capacity "immediately exposed" by Hinojosa's win at roughly 13 gigawatts, enough to power about 10 million homes at any moment. Jefferies estimated 13.3 gigawatts of announced Texas projects are at risk. Nationally, Bloomberg's read is that Democratic wins from Texas to Ohio and Pennsylvania expose half of all AI data center proposals in the country to more scrutiny. Markets moved before the counts finished: AI data center shares fell and renewables rose, Reuters reported, and CNBC tied the slide directly to the Texas governor's race.
The binding constraint on AI capacity just changed character, and that is a bigger deal than any single project. For two years the bottleneck has been physical — transformers, turbines, substations, water, electricians. Those constraints are expensive, but they are legible: you can price a five-year lead time on a gas turbine and schedule around it. A county ballot measure is not legible. It arrives after the land is bought and the power is contracted, it can arrive in a village of 400 people, and it cannot be outbid. A hyperscaler can out-negotiate a utility. It cannot out-vote a majority.
There is a second effect that the sell-off mostly missed. What got cheaper on Tuesday was not compute; it was the option to build more of it. If 13 GW of Texas capacity, much of it still in early stages, is delayed rather than cancelled, the near-term supply of installed capacity tightens while the long-term supply of announced capacity stays where it was. Scarce installed compute and slower uninstalled buildout is a mix that favors whoever already has powered shells — the incumbents with operating campuses — and penalizes labs competing for new capacity. Ben Thompson made a version of this argument on Stratechery: for hyperscalers stuck in a capex arms race none of them wanted to slow, being forced to slow by voters may be the outcome they could not choose themselves.
The honest counter-argument is that the exit polls cut both ways, and everyone spent Wednesday claiming them. NBC's national exit poll found 56% of voters called AI and data centers a "crucial or major factor" in their congressional vote, per the Financial Times, but only 23% called it crucial — and The Decoder notes AI ranked last among the five issues the poll asked about. Adam Kovacevich of the pro-tech Chamber of Progress spent the morning pointing at exactly that. Both readings are true at different altitudes: data centers were not why most Americans voted, and where they were on the ballot, they lost. Local salience decides permits. National salience decides nothing.
The second reason to discount the panic: Hinojosa's four conditions are not a ban, and most of them are already industry practice in public. Operators that drop NDAs and pay for grid upgrades can still get approved under her framework — the state's House and Senate also remain Republican, which limits how far an executive freeze can go. The realistic outcome for Texas is delay and renegotiation, not demolition. But a project delayed two years misses the 2028 capacity window, which is the same thing as a cancellation for anyone counting gigawatts against a training roadmap.
There is also the money question nobody has answered. Silicon Valley spent heavily to prevent this and lost anyway: Musk's $474 million, roughly $250 million from crypto and AI-aligned super PACs, $94.4 million tied to Andreessen Horowitz, all per Molly White's Tech Influence Watch tally cited by 404 Media. Campaign spending is the industry's cheapest lever, and it did not work. The expensive lever — litigation, local lobbying, paying for grid upgrades — is still available, but it does not scale to 40 ballot measures a cycle.
Ohio is the template worth watching, because it was supposed to be safe. Sherrod Brown won his Senate race by five points running on reining in data centers, in a state that has trended deep red, and Vivek Ramaswamy lost to Amy Acton, a public health official who supports moratoriums. Kathy Hochul won reelection in New York after enacting the country's first statewide moratorium on new hyperscale data centers in July. That is three different offices, three different parties' terrain, one position.
What to watch: whether Hinojosa moves by executive action or waits for the legislature, since the difference decides whether Texas projects stall in January or in 2027; the capex guidance hyperscalers give on Q4 earnings calls in the next few weeks, which will show whether they are re-routing capacity or re-pricing it; and how many of Tuesday's measures end up in court, since Urbana's suit is the first of what will be a long line. One measure out of 40 went the industry's way. That is not a backlash to wait out; that is a map to build around.
If your county put a data center on the ballot, would you have voted for it? Write to us via our contact page.
