Communities aren’t rejecting AI. They’re rejecting the bait-and-switch — build fast, disclose later, pay penalties as a line item. I’ve spent enough time inside enterprise infrastructure to know the difference between resistance that kills innovation and resistance that forces it to grow up. Today, most of what’s happening globally is the second kind. Not all of it.
A farmer in Vineland, New Jersey, was photographing a neighbour’s golf course last February and caught something odd: rows of unmarked generator units located on land where a Microsoft-linked data centre was under construction. He flagged it at a town hall. Six months passed before New Jersey regulators showed up. They found 62 unpermitted natural gas generators running without a single required air permit, powering a facility built for a $17 billion AI infrastructure deal. The fine assessed: $1.07 million — the largest data centre enforcement action in the state’s history, and, by resident Matt Williams‘s account, “unsurprising and long overdue.”
I want to be forthright about where I sit here, because most coverage views data centre opposition as a single, undifferentiated wave of NIMBYism standing between the world and its AI future. It isn’t. Some of it is that. Most of what I’m seeing is something else: a rational, increasingly organised response to an industry that has spent three years building faster than it discloses, and treating the gap between the two as an acceptable cost of doing business.
The Fighting on Multiple Fronts
Governments are the slowest-moving front, and the most consequential once they move. New Jersey didn’t just fine DataOne — it passed a law ending data centre tax credits, cutting developers off from a fund that offered $250 million in incentives. Virginia fined a Microsoft facility $2.5 million for exceeding emissions limits the week before New Jersey’s action. Stokes County, North Carolina, is fighting its rezoning decision in court over a data centre complex built without assessing impact on Saura tribal history and enslaved-descendant cemeteries. This isn’t a coordinated campaign. It’s dozens of separate jurisdictions reaching the same conclusion at the same moment: the current disclosure and permitting regime wasn’t built for facilities drawing more power than the towns around them.
Legal challenges are the fastest-moving front, and the messiest. Litigation spans procedural due process, open records law, environmental protection, common law nuisance, historic preservation, vested rights, trade secrets, and constitutional claims — any one of which can delay or kill a project on its own. A Texas developer filed a $100 million lawsuit against the state’s first county-level data centre moratorium, arguing it lacked legal authority. Sustain SJ, the group formed to oppose the Vineland facility, is pushing for a statewide moratorium until transparency requirements improve — more than 80 data centres are planned or under construction in New Jersey alone. Data Centre Watch counted $68 billion worth of projects blocked or delayed in a single quarter across 11 US states.
Innovators are the quietest front, and arguably the most interesting. Every fine, every moratorium, every year-long permitting fight is a sign to the market: build, or keep losing time and capital to a fight you can avoid. That signal is starting to land.
Does Resistance Help or Hurt Innovation?
Let’s push back on the perspective I see most often in the industry. The claim is some version of: community opposition slows AI progress, therefore it’s bad for innovation. That’s true if you define innovation narrowly, as raw compute capacity coming online as fast as possible. Widen the lens and the picture flips.
Where resistance stifles progress: the UK’s Loughton supercomputer, announced by the Starmer government in 2025 as the country’s “largest sovereign AI datacentre,” faces a delay into the early-to-mid-2030s — not because of community opposition, but because UK Power Networks can’t connect the site to the grid until then. That’s a different failure mode, and it’s the one I’d flag as a real innovation cost: a national AI strategy built around a physical resource — grid capacity — that nobody planned for before announcing the project. Nscale is seeking a $35 billion flotation while its flagship site sits waiting on infrastructure it doesn’t control. That’s not communities failing innovation. That’s planning failing communities and innovators at the same time.
Where resistance is forcing better products: the DataOne case is the clearest example I’ve seen of accountability working as designed. Community members used satellite imagery and drone photography — democratised surveillance tools — to document a violation regulators missed for months. The company’s spokesperson said it’s transitioning to lower-emission fuel cells. That’s not a company being punished into paralysis. That’s a company being caught, fined, and pushed toward a better power architecture it might not have adopted on its timeline. If the alternative to that pressure is 62 unpermitted gas generators running for years, I’d call the pressure a feature, not a bug.
London’s Brick Lane fight cuts a third way. Bangladeshi community members are fighting a data centre proposed for a disused brewery, arguing it threatens heritage and displaces needed social housing — the redevelopment includes only 44 flats, six of them affordable. The UK government approved it anyway, citing “significant weight” for AI capacity needs. I don’t think that stifles innovation at all. I think it’s a legitimate argument about whose needs get “significant weight” in a housing crisis, dressed up by the developer as a tech-versus-NIMBY story because it’s easier to win.
Employing Alternative Models
I don’t think the honest answer here is “build less.” Compute demand is real, and pretending otherwise doesn’t help anyone making infrastructure decisions. The honest answer is that the current default model — announce fast, permit as you go, disclose when caught — is a worse business model than most operators currently admit, and better alternatives already exist in pieces:
Community investment as a permitting condition, not a PR afterthought. New York State has recommended AI data centres pay $1 million in community investment per megawatt — a serious number that resets the negotiation from “please don’t oppose us” to “here’s what capacity costs the people living next to it.” That figure alone would have changed the DataOne calculus; a $1.07 million fine on a 300-megawatt facility is closer to a rounding error than a deterrent.
Front-loading transparency instead of fighting disclosure requests. Every legal theory used against developers—open records, environmental review, historic preservation—becomes a weaker weapon when a developer has already published the information residents are suing to obtain. The operators TF has covered engaging communities early, before ground breaks, face fewer lawsuits later. That’s not a coincidence; it’s the entire dispute resolving itself before it becomes a dispute.
Grid-honest planning, not announcement-first politics. The Loughton failure is a template for what not to do: announce a flagship national project before confirming the physical infrastructure exists to support it. Any government or company serious about scaling AI compute needs to run the grid-connection timeline first and the press release second — not the reverse.
Distributed and alternative compute architectures. As TF covered in Google’s First Orbital Data Centre Launches 1 October, the industry is already exploring varied models — orbital compute, quantum-adjacent approaches, and smaller distributed facilities rather than single mega-sites drawing more power than the towns around them. None of these solves the near-term capacity crunch. All of them suggest the industry itself doesn’t believe the current model is the only viable path forward, whatever its public messaging implies.
TF Summary: What’s Next
I’ve spent enough years inside enterprise cloud and platform infrastructure to know the difference between a community killing a project it doesn’t understand, and a community catching a company that built first and planned to disclose later, if ever. Most of what I’m tracking is the second kind. The DataOne case, the Virginia fine, the North Carolina rezoning fight — none of these are communities rejecting AI. They’re communities rejecting a specific, repeated pattern: build the facility, run it non-compliant, treat the eventual fine as a cost of doing business, and let residents do the regulator’s job of catching you.
MY FORECAST: Expect the community investment model New York has proposed to become the industry standard within two years, not because operators embrace it voluntarily, but because the litigation and moratorium wave TF has tracked across at least 11 states makes the current build-first approach more expensive than the alternative. The Loughton failure will be cited in every subsequent national AI infrastructure announcement — no government wants to be the next headline about a flagship supercomputer waiting on a grid connection nobody confirmed in advance. And the innovators who figure out how to build smaller, more distributed, more transparent compute — rather than fighting every community that asks reasonable questions about power and water — will spend less time in court than the operators still betting the old model holds.
Related Stories
- Google’s First Orbital Data Centre Launches 1 October
- Andrew Bailey Warns G20 of AI Bubble Financial Stability Risk
- RAMageddon Hits MacBook Air Supply as Chip Shortage Deepens