Voters Just Blocked 40 of 41 Data Center Projects. Permission Is a Real Risk.

Forty-one data center related measures went before voters in the November 2026 midterms, spread across roughly ten states including California, Florida, Illinois, Kansas, Maine, Nevada, New Jersey, Ohio, and Wisconsin. Coverage of the results converged on the same count: the industry won one of them, in Harwood, North Dakota. In the rest, voters either banned new data center construction outright or halted facilities already in the pipeline.

Two details make this more than a local zoning story.

First, more than half of the measures were put on the ballot by citizen initiative rather than by elected officials. This was not a case of politicians getting ahead of their constituents. It was residents organizing the paperwork themselves.

Second, the spending. Reporting indicates technology companies and allied groups outspent local campaigns by something on the order of 100 to 1, and still lost nearly everywhere. A Gallup survey from March 2026 had already shown about 70 percent of Americans opposed to new AI data centers in their own neighborhoods. The money was spent against a preference that was already set.

Most readers are not building data centers. The transferable lesson is not about servers. It is about a category of risk that early teams routinely leave off the board deck: the risk that your business requires permission from people who are not your customers and who get nothing out of saying yes.

Call it permission risk. It shows up any time your model depends on a physical footprint, a license, a shared resource, or a rule that a third party controls. Delivery robots on sidewalks. Short-term rentals. Micro-mobility. Battery storage. Vertical farms. Clinics. Childcare. Drone logistics. Anything that touches water, power, noise, traffic, or land use.

The failure mode is specific and common. A team validates demand carefully, builds a product customers want, raises on that traction, and then discovers that the binding constraint was never demand. It was a planning board, a utility interconnection queue, a state licensing body, or four hundred neighbors who turned up to a Tuesday evening hearing.

Demand validation and permission validation are separate exercises. Teams that conflate them tend to find out in the quarter they can least afford it.

The 100 to 1 spending gap is the single most instructive number in the whole result. It suggests that above some threshold, additional spending on persuasion stops helping and starts confirming the opposition's story.

The mechanism is not mysterious. When a well funded outside party floods a small community with advertising, the spending itself becomes evidence of asymmetry. Residents conclude that someone with a lot to gain is working hard to convince them of something, which raises rather than lowers suspicion. Local credibility does not transfer from a budget.

What tends to work instead is slower and less scalable: showing up early, before a proposal is filed, with specifics about what the community gets and what it will have to tolerate. Concrete commitments on noise limits, water usage, traffic routing, local hiring, and tax contributions. A named person who answers the phone. A willingness to change the plan when the first version is wrong.

None of that fits neatly into a growth model. That is the point. Businesses with permission risk cannot be scaled purely with capital, and founders who treat consent as a line item to be purchased tend to learn this expensively.

A useful exercise for any team whose model touches the physical world. It takes a few hours and it should be done before the first site, not after.

List every party whose approval you need, formally or informally. Not just regulators. Include utilities, neighbors, local business associations, unions where relevant, and the specific committees that hold hearings.

For each one, write down what they get if they say yes, what they absorb if they say yes, and what happens to them personally if the project goes badly. If the second column is heavier than the first for an important group, you have a problem that no amount of product quality fixes.

Then find the precedent. Someone has tried something similar within a hundred miles of your first site. Find out what happened, read the hearing minutes if they are public, and talk to whoever led the opposition. That conversation is uncomfortable and it is the highest-value hour in the whole process.

For founders working through this kind of structural risk assessment for the first time, Foundra's planning framework prompts for exactly these questions early, while the answers can still change the plan instead of just explaining a loss.

Harwood, North Dakota is the exception, and exceptions are worth examining more carefully than the pattern.

The broader reporting does not credit a clever campaign. Smaller communities with existing industrial land use, available power, fewer immediately adjacent residents, and a visible local tax need behave differently from suburban districts where the primary local asset is quiet. The outcome tracked the pre-existing conditions more than the persuasion.

The founder translation: site selection is a strategy decision, not a logistics decision. The most common error is optimizing a location for cost per unit and treating community acceptance as something to be handled later by a government relations hire. Often it is cheaper, in total, to pay more for a site where the answer is likely to be yes.

This generalizes beyond land. The same logic applies to choosing which state to launch a regulated service in, which hospital system to pilot with, or which city to run delivery in first. Pick where the structural answer is already leaning your way, prove the model, and use that record when you walk into the harder rooms.

Permission runs on calendars you do not control, and this is where otherwise careful financial models break.

Hearings get scheduled monthly, not weekly. Appeals add quarters. Utility interconnection studies can take a year or more. A single continuance can push a project past the end of your runway, and the people granting the continuance have no stake in your burn rate.

Build your plan around this rather than hoping. Three practices help. Keep a parallel pipeline, so one blocked site does not stop the company. Sequence spending so that major capital commitments come after the binding approval, not before it. And state your assumed timeline explicitly to investors, including the version where it takes twice as long, because the alternative is explaining a surprise later.

The 2026 results also added a new variable. Measures that reach a ballot introduce an election calendar into your project timeline. That is a different kind of delay from a permit review, because it has a fixed date and an unpredictable outcome, and it cannot be accelerated by being more prepared.

Even a company with no physical presence has a version of this problem, and the data center vote is a reminder to go looking for it.

The analogous question: who can switch you off? App store policy. A platform's API terms. A payment processor's risk team. A single cloud provider's acceptable use policy. A professional licensing body that decides your tool constitutes practicing their profession. None of these parties are your customers, all of them can end your business, and most founders have never written their names down.

The exercise is the same as the land-use one. List them. Note what each gets from your existence and what each absorbs. Identify which single party could cause the most damage fastest, and ask what a reasonable hedge costs today.

The broader point of the midterm result is not that building physical infrastructure is harder than it used to be, although it is. It is that public tolerance is a resource like capital or talent, it can be depleted, and when it runs out, it does not respond to spending. That is worth understanding before your business depends on it.

How many data center measures were on the ballot in 2026? Reporting identified 41 data center related measures across roughly ten states. The industry prevailed in one, in Harwood, North Dakota, while voters curbed or blocked development in the rest.

Does this mean data centers will stop being built? No. It means siting them in communities that get a vote has become considerably harder and slower. Development is likely to concentrate where local conditions and incentives already favor it.

What is permission risk? The risk that your business requires approval from parties who are not your customers and receive little benefit from saying yes. It includes regulators, utilities, licensing bodies, platforms, and neighbors.

When should a startup start community engagement? Before filing anything. Once a proposal is public, the conversation becomes a defense of a fixed plan rather than a collaboration on an open one, and positions harden quickly.

Why did heavy spending fail to change the outcome? Large outside spending in small communities tends to signal asymmetry of interest, which reinforces suspicion. Local credibility is built through presence and specific commitments rather than advertising volume.

My startup is software only. Is any of this relevant? Yes. Identify every third party that could switch you off, including app stores, platforms, processors, and licensing bodies, and treat that list with the same seriousness a hardware company gives permitting.

1. Voters rebuff data centers in every ballot measure except one during the 2026 midterms (CNBC)
2. Here Is Where Voters Are Rejecting Data Centers On Midterm Ballots, Full List (Forbes)
3. How the US midterm elections reflected resistance to data centres (Al Jazeera)
4. Voters Reject Data Centers in 2026 Midterm Elections (Capital B News)
5. After the midterms, things are looking rocky for Elon Musk and data centers (TechCrunch)