This is not a piece about who won. We are a firm that builds things, our clients vote in every direction, and a design studio handing out election analysis would be both unwelcome and unqualified. So nothing below celebrates or mourns an outcome, and nothing below assigns credit or blame.
What we are qualified to talk about is planning. And on Tuesday, a set of facts landed that changes the planning assumption for any business whose operations now depend on AI — which is most of them. Those facts are legible without taking a side, and ignoring them is expensive in a way that has nothing to do with partisanship.
AI was on the ballot literally, not just rhetorically
Start with the part that required no interpretation, because it was printed on ballots.
Ballotpedia counted 41 data-center-related measures across ten states voted on November 3, with Ohio alone accounting for 21 of them. Twenty-eight of the 41 were aimed at limiting development — size-based or total bans, six-month moratoriums, new zoning rules, voter-approval requirements. Thirteen were aimed at facilitating it.
Semafor reports that voters in more than 30 localities across nine states approved moratoriums, bans, or other restrictions, several in direct response to specific projects already underway. Individual examples are concrete: Edgewater, Florida wrote a data-center ban into its city charter; Piqua, Ohio approved a ban on large data centers; Ashville, Ohio repealed a village agreement with a developer.
The salience shows up in survey work too. The NBC News exit poll found 56% of voters called AI and data centers a crucial or major factor in their congressional vote — 23% crucial, 33% major. A POLITICO poll conducted October 16–19 found half of Americans oppose a data center built within three miles of their home, up from 28% in January.
Whatever you make of the politics, that is a fast-moving public opinion curve on a specific piece of physical infrastructure. And it is not a partisan artifact: a Reuters/Ipsos poll of 4,506 adults fielded September 30–October 5 found 69% want stricter government regulation of AI, including 80% of Democrats, 61% of Republicans, and 55% of independents. There is no coalition in that data that wants fewer rules.
The near-term business issue is compute and power, not policy
Here is where it stops being a political story and becomes a line item.
Every AI capability your company rents runs on a building somewhere, drawing power from a grid. Residential electricity prices have risen more than 10% since 2024, faster in several states that are also data-center growth corridors, and the International Energy Agency's figure for a typical AI data center is consumption equivalent to roughly 100,000 homes.
So you now have local restriction stacking on top of rising power cost, while demand for compute keeps climbing. We are not going to forecast a price. We will say the direction is unhelpful, and that the planning posture that assumes AI inference gets cheaper and more abundant every quarter forever is a posture, not a fact.
Practically, that argues for the thing we already argue for: don't build a business process whose economics only work at a vendor's current promotional rate, and don't build one that breaks if a model gets deprecated, rate-limited, or repriced. Keep your own data, your own documentation, and your own record of how the work gets done, so that swapping the engine underneath is an inconvenience rather than a rebuild. We wrote the longer version of that case in The moat is what you own.
The federal picture: watch the preemption fight
There is a lot of confident commentary right now about a sweeping federal AI law arriving shortly. We would treat that as a forecast rather than a fact, and a contested one.
The useful thing to anchor on is the bill that actually exists. The FRONTIER Act — H.R. 9925, the Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting Act — was introduced July 23, 2026 by Rep. Jay Obernolte with bipartisan original cosponsors including Rep. Scott Franklin and Rep. Lori Trahan. It is the clearest published statement of what a federal frontier-AI regime would look like:
Tiered obligations by developer size. It targets the handful of companies training the most capable models rather than every company using AI, anchored to a compute threshold on the order of 10²⁶ operations, adjustable by rule.
Published risk frameworks and model documentation. Covered developers would have to write down and publish how they assess catastrophic risk, secure model weights, and make deployment decisions.
Independent verification and incident reporting. The largest developers would retain licensed third-party evaluators; safety incidents would be reported, with state attorneys general able to opt in to receive them.
Federal preemption, with carve-outs. It would bar states from imposing new substantive obligations inside the covered transparency, audit, and reporting domain, while preserving generally applicable law, use and deployment rules, protections for minors, and state procurement.
As of this writing the bill has been referred to committee with no floor vote. Its own analysts flag the path as uncertain — overlapping committee jurisdiction, House–Senate differences, and an unresolved argument over preemption. Any reporting that treats passage, or the shape of a successor bill, as settled is predicting, not reporting.
That preemption clause is the part worth watching, and it is the part that matters to you. The fight is not really regulation versus no regulation. It is one national rule set versus fifty-plus state and local ones, and the honest answer is that nobody currently knows which way it resolves. Brendan Steinhauser of the Alliance for Secure AI expects, in his words, "a big deluge of bills at the state and federal level."
Notice that both branches lead to the same operational outcome. A federal framework means disclosure and documentation requirements written once. A failed federal framework means those requirements written separately by many jurisdictions, which is more work, not less. There is no live scenario in which the paperwork gets lighter.
This is the same through-line, arriving from a new direction
We have written this argument before from the technical side and the product side, and it keeps showing up because it is structural rather than topical.
In September we covered model-level watermarking and C2PA content credentials — the fact that AI-generated text and files increasingly carry signed provenance, and that the exposure is not that you used AI, it is the gap between what you said about your process and what the record shows. In the piece on visual agents we made the case that an agent speaking for your company is a representative, and needs plain disclosure and a named human behind it.
Tuesday added a third source of the same pressure, from the least technical direction possible: ordinary voters, in municipal elections, about a building. Provenance infrastructure, interface honesty, and local politics are not related topics. They are three instruments reading the same trend. Disclosure, human accountability, and owning your own data are moving from differentiators to table stakes.
We say this as a firm with an obvious interest in it being true, so discount accordingly — but we also operate that way already, and said so before it was required. Pixelworkers uses AI across research, drafting, analysis, and production; we have disclosed that publicly for as long as we have done it; a human with domain expertise sets direction and is accountable for every result; and the work ships in the open where anyone can check it. Regulation of this shape does not threaten firms whose process survives being described accurately. It just makes the difference visible.
What to do this quarter
Four things. None require a lawyer, a lobbyist, or an opinion about the election.
1. Write down and publish your AI standard. One page: where AI is used, where a human reviews, who signs off, what never ships without a person. Published voluntarily, it reads as confidence. Produced under questioning, the identical document reads as an admission. The content is the same; the timing is the asset.
2. Keep a named human accountable for every output. Not a review process in the abstract — a person, by name, who owns the result and can be escalated to. Every disclosure regime proposed so far, state or federal, eventually asks who was responsible. Have an answer that is a name.
3. Hold your own data and your own documentation. Customer records, conversation logs, process documentation, prompts and workflows that encode how your business actually works — stored where you control them, exportable, not stranded inside whichever vendor you happened to pick in 2026. This is good practice regardless; it becomes necessary the moment someone asks you to produce a record.
4. Don't build a plan that requires deregulation. This is the one we would underline. If a workflow, a margin, or a roadmap only pencils out in a world with fewer AI rules and cheaper compute, you are not running a business strategy, you are holding a position on a policy outcome. Nothing in the available evidence — not the ballot measures, not the polling across either party, not the bills already filed — supports that position.
The part worth sitting with
For about three years, AI policy was something that happened to AI companies. It was a conversation among labs, regulators, and the trade press, and most businesses could reasonably treat it as somebody else's problem while they got on with using the tools.
That ended quietly on Tuesday, in township elections, over zoning. The constituency for AI rules turned out not to be policy specialists. It was people looking at a utility bill and a construction site.
Which means the requirements, when they arrive, will be shaped less by technical sophistication than by a general demand to know what is going on — who made this, with what, and who stands behind it. Those are not hard questions. They are only hard to answer retroactively.
So answer them now, while answering is still voluntary and therefore worth something.
Sources: Ballotpedia — data-center ballot measures, November 3, 2026; Semafor — US voters reject data centers in ballot measures; H.R. 9925, FRONTIER Act (congress.gov) and sponsor summary; Reuters/Ipsos topline, fielded September 30 – October 5, 2026; POWER Magazine — data centers, affordability, and the ballot box (POLITICO/PF Global poll, October 16–19, 2026); NBC News exit poll, November 4, 2026, as reported by TIME (also the source of the Steinhauser quotation). Associated Press — power bills and data centers in the US midterms (October 31, 2026), for electricity price increases and the IEA data-center consumption figure. Accessed November 4, 2026. Characterizations of the FRONTIER Act's prospects, and of what Congress will do next, are forecasts by the cited parties, not settled outcomes.
