The poll measures how likely local opposition is, and that likelihood is what now shapes siting decisions more than any single permit does. Support for a nearby data center dropped 28 points in twelve months [1], and the ratio of opposition to support moved from roughly even to about five to one [2], which raises the odds that any announced campus becomes a contested one. Contested is where the money sits: Heatmap puts the cancellation rate for disputed developments near 50 percent in the first half of 2026, against 20 percent in late 2024 [4], a 2.5-fold increase in the penalty for losing the room [3]. On the same tally, more than 300 projects have been either killed or pushed into open dispute this year [4].
The board-deck rebuttal runs the other way, and it is a real one. More than 90 percent of US counties still had no significant constraints at the end of July [6], meaning fewer than one in ten did [6], and New York and Pennsylvania, whose new rules Vox judges the most substantial of the recent batch, together host about 6.5 percent of American data centers according to Data Center Map [10], leaving roughly 93.5 percent of the installed base elsewhere [5]. The slide is incomplete on two counts. It counts counties rather than counties that can actually deliver power to a hyperscale campus, and the source record does not tell us how many of those are left. It is also a snapshot of a stock being drawn down, since most of the 500-plus restrictive jurisdictions adopted their measures since January [3].
Texas's restriction is a power-sourcing filter that lets self-generation projects skip the screening. The state's pause does not halt construction; it adds a more thorough screening step when a project applies for electricity from the state grid, and projects that supply all of their own power from on-site natural gas are exempt [9]. That filter sorts on power sourcing rather than on location, and a developer who values schedule certainty has an obvious design answer. My read is that the exemption moves the argument from a state grid process to the local land-use bodies where the 15 percent figure lives [1], which is the harder venue rather than the easier one.
Polls do not pour concrete, and Vox's own conclusion makes that case directly: absent a congressional moratorium, AI firms will almost certainly keep finding jurisdictions willing to tolerate campuses in exchange for revenue and incentives, so the buildout can be slowed and relocated but probably not killed short of a collapse in compute demand [11]. That is right at the level of national capacity and beside the point at the level of the asset. Aggregate compute can keep compounding while a particular site dies at a county meeting, and the party carrying that loss is the developer who picked the county.
So the tradeoff is speed against consent, and the two horizons price it differently. A developer who relocates to a permissive jurisdiction keeps the schedule and spends no political capital, which works until that jurisdiction changes its mind. A developer who negotiates pays in months and in disclosure. Over a decade the only lever the record identifies as able to bind the whole buildout is a federal one, and Vox treats that federal moratorium as hypothetical [11]. Between those two horizons, a plan approved this quarter is worth more if it names three candidate sites than if it names one, and that optionality has become a line item rather than a contingency.
