Seventy to seventy-five percent of Americans don’t want a large data center built near their home. That’s a Gallup number, not a vibe, and more than half of respondents go further and support a national moratorium on building them. For an industry that spends its days talking about scaling laws and compute curves, that’s a wall nobody modeled.
I build bots for a living. My work depends on cheap inference, available GPUs, and someone, somewhere, pouring concrete. So I have skin in this game, and I’d rather look at the problem honestly than pretend it’s a messaging issue.
Secrecy as a default setting
The specific question making rounds right now is whether the industry could win over opponents by dropping the non-disclosure agreements it uses while negotiating with local governments over data center projects. It’s a narrow ask with a wide blast radius. NDAs in municipal negotiations mean a town council can be deep in talks about power draw, water use, tax abatements, and land rezoning while residents are told nothing until the deal is effectively done.
If you’ve ever shipped software in a team that hid its roadmap from the people who had to maintain it, you know how this ends. The decision might have been fine. Nobody believes it was, because nobody could check.
The reporting on this doesn’t land on a verdict. There’s no clean answer on whether killing NDAs would actually move opposition numbers. That ambiguity is the interesting part, and I think it’s worth sitting with instead of resolving prematurely.
Why I suspect transparency alone won’t close the gap
Here’s my read as someone who debugs systems for a living. NDAs are a symptom. The disease is that the people absorbing the costs of a facility and the people capturing its value are usually not the same people.
Open the books and you get transparency about a deal that residents may still dislike. Transparency is necessary. It is not persuasion. A town that learns the exact megawatt figure it was going to get anyway has more information and the same complaint.
Michael Spencer’s framing of an accelerating AI backlash tracks what I’m seeing. In 2025, general sentiment toward AI slid even as agentic AI failed to deliver for a lot of businesses. In 2026, that soured mood found a physical target. Models are abstract. A substation is not.
The politics have already caught up
Pennsylvania is the tell. Governor Josh Shapiro signed an executive order imposing strict standards on data center development. That’s not a protest sign, that’s state machinery. When a governor acts, the window for voluntary industry gestures narrows considerably, because the rules are being written whether or not anyone volunteers.
There’s also a real fight over facts. The Goldwater Institute ran a piece in April 2026 titled “The Data Center Debate: Fact vs. Fiction,” which tells you the dispute over what’s true has become its own front. Meanwhile Amazon reportedly rewired its data centers following a power warning from Mississippi Power. Grid operators are sending signals, and companies are responding to them.
What this means if you build on top of all this
If you’re writing agents, fine-tuning models, or running anything with a token meter attached, this is your supply chain. Some practical thoughts:
- Treat compute as a politically contingent resource, not a utility. Siting fights can slow capacity, and slow capacity shows up as price and availability.
- Build for efficiency now rather than later. Smaller models, caching, batching, and honest evaluation of whether a task needs a frontier model at all. Lower draw is a defensible position, technically and publicly.
- Don’t oversell what you ship. Every inflated agent demo contributes to the sentiment decline that’s now showing up at zoning hearings.
- Know where your inference physically runs. If you’re serving customers in a region with active opposition, that’s operational risk worth tracking.
My honest guess
Dropping NDAs is the right move and insufficient on its own. Right, because secrecy in public negotiations converts ordinary skepticism into suspicion of a cover-up, and that’s a much more expensive problem. Insufficient, because the opposition numbers suggest people have a substantive objection, not an informational one.
What would actually shift things is harder and less flattering: binding commitments on power and water that residents can verify, cost structures that don’t push rate increases onto households, and a willingness to lose some siting fights. That’s a different kind of engineering than we’re used to, with a feedback loop measured in election cycles.
The industry is good at optimizing for benchmarks. This problem has no leaderboard, and the people scoring it live next door.
🕒 Published: