Years before “AI data centers” became a phrase people argued about at town halls, communities were already learning the same hard lessons: large-scale computing doesn’t arrive as a neutral technology. It arrives with land use fights, grid capacity questions, water and cooling constraints, tax and zoning negotiations, and—often—an uncomfortable mismatch between who benefits and who bears the costs.
That’s why the current wave of resistance to AI expansion doesn’t feel like a sudden backlash. It feels like a replay. The players change—sometimes the company name does, sometimes the stated purpose shifts from cloud services to machine learning—but the underlying friction is familiar. And in many places, the groundwork was laid long before the AI boom made the stakes feel urgent.
A useful place to start is 2015, when Apple announced plans for a major data center in Athenry, Ireland. The project was framed as infrastructure for services across Europe, including iTunes, iMessage, and Siri. It was also, by any measure, a big footprint: a roughly $1 billion investment tied to a site that would be hundreds of acres in scale. At the time, the public conversation wasn’t yet dominated by generative AI. But the core issue—what happens when a global tech company builds energy-hungry facilities in a local landscape—was already front and center.
The Athenry proposal didn’t just raise questions about construction. It raised questions about governance. Who gets to decide what kind of development is appropriate? How do local planning processes weigh economic promises against environmental and infrastructure burdens? What does “capacity” mean when it’s not just a technical term but a lived reality for residents who experience outages, traffic, noise, and long-term changes to their region?
Those early debates mattered because they established patterns that later communities would recognize instantly. They showed that data center projects can become political objects, not merely business decisions. They demonstrated that opposition can be organized around more than one concern—environmental impact, visual and land-use changes, strain on utilities, and the perceived lack of transparency about long-term operations. And they revealed something else: even when a project is eventually approved, the conflict can leave behind a blueprint for future fights.
That blueprint is now being reused across jurisdictions as AI demand accelerates buildouts. The difference is that today’s projects are often justified with a different urgency. Instead of “cloud services,” the pitch is “AI compute.” Instead of incremental growth, the timeline can feel compressed. Instead of a single facility, the plan may involve multiple campuses or phased expansions designed to keep pace with rapidly changing demand.
But the questions communities ask are strikingly consistent:
Where will these facilities be built?
Can local power systems handle the demand?
How will communities plan for the impacts?
Those questions sound straightforward, but they’re difficult in practice because data centers sit at the intersection of several systems that don’t coordinate neatly. Power grids are planned on long horizons, while tech companies can scale compute needs quickly. Zoning and permitting processes are local and procedural, while data center economics are global and often negotiated through complex corporate structures. Environmental assessments can be thorough, but they may not fully capture the operational reality of decades-long energy consumption and the evolving efficiency of hardware.
In other words, the fight isn’t only about whether a data center is “good” or “bad.” It’s about whether the surrounding systems—energy, water, transportation, land use, and governance—are ready for the kind of load that modern compute clusters represent.
The early Apple example in Ireland is instructive not because it predicts every outcome, but because it illustrates how large-scale infrastructure can collide with local concerns before the public has a shared vocabulary for the problem. In 2015, “AI data center” wasn’t the headline. Yet the project still forced a confrontation with the same fundamentals: land, power, and community impact.
That matters because it reframes the current moment. If today’s resistance is treated as a sudden reaction to AI, it becomes easy to dismiss as misinformation, fearmongering, or NIMBYism. But if you look at the longer arc, you see something more structural: communities have been negotiating the terms of digital infrastructure for years, and AI has simply intensified the pressure.
One reason the resistance is gaining traction now is that the scale of demand is no longer theoretical. AI workloads can require massive amounts of electricity, and the buildout of supporting infrastructure—substations, transmission upgrades, backup generation, cooling systems—can take time. Even when individual facilities are designed with efficiency in mind, the aggregate effect of many facilities can overwhelm local planning assumptions.
This is where the “where” question becomes inseparable from the “power” question. Data centers can be sited near existing power capacity, but that capacity is not infinite. In some regions, the grid may be able to supply the initial load, but not the future expansions that companies request. In others, the grid may require upgrades that are expensive, slow, and politically contentious. Residents may not experience the upgrade process as a technical necessity; they may experience it as a disruption—construction noise, new lines, altered landscapes, and uncertainty about timelines.
And then there’s the “how” question: how do communities plan for impacts when the impacts are partly unknown at the time of approval? Data center operations evolve. Hardware refresh cycles change energy profiles. Cooling strategies can shift. Workloads can vary dramatically depending on customer demand. Even if a project’s initial footprint is defined, its long-term trajectory may not be fully locked down.
That uncertainty is a major driver of opposition. People aren’t only asking, “Will this facility be built?” They’re asking, “What exactly are we agreeing to, and what happens if the plan changes?”
In many places, the answer has been: not enough clarity. Or at least, not enough clarity early enough to satisfy residents. That’s why transparency has become a recurring theme in these conflicts. Communities want detailed information about energy sourcing, backup power usage, emissions, water consumption, and the expected timeline for grid upgrades. They also want clarity about who is responsible for mitigation if impacts exceed projections.
The unique take in the current moment is that AI has turned these questions into a broader public conversation. When data centers were built primarily to support general cloud services, the debate could remain niche—technical, local, and sometimes limited to those directly affected. With AI, the stakes feel more immediate to a wider audience. People understand that AI requires compute. They may not know the engineering details, but they grasp the basic idea that more AI means more electricity and more infrastructure.
That shared understanding changes the political dynamics. It makes it easier for opponents to recruit allies, harder for proponents to frame the project as purely beneficial and inevitable, and more likely that elected officials will face pressure to respond.
Still, it’s important to avoid turning the story into a simple morality play. Proponents of data centers often argue that they enable economic growth, support jobs, and provide essential digital services. They may also point to efficiency improvements, renewable energy procurement, and modern designs that reduce certain environmental impacts compared to older facilities.
Opponents, meanwhile, often argue that the benefits are unevenly distributed and that the costs—especially energy strain and local environmental effects—are borne locally while profits and decision-making remain distant. They may also argue that the pace of buildout outstrips the ability of communities to adapt.
Both sides can be partially right, which is precisely why the conflict persists. The real question becomes: can the planning system evolve fast enough to manage the scale and speed of modern compute infrastructure?
The Athenry case shows how early conflicts can shape later ones. When a project like that is proposed, it forces a community to confront the mechanics of approval: what documents are required, what hearings occur, what mitigation measures are offered, and what enforcement exists after construction. Even when the final outcome is not what opponents want, the process teaches them how to organize, what arguments resonate, and which institutions hold leverage.
Over time, that learning accumulates. It becomes institutional knowledge among local activists, planners, and policymakers. It becomes a set of expectations about what should be demanded from future proposals. And it becomes a narrative that can be repeated: “We’ve seen this before.”
Now, as AI-driven buildouts accelerate, that narrative is spreading. Communities that once focused on general data center concerns are increasingly framing their objections in terms of AI’s specific energy demands and the risk of locking in infrastructure that may not align with long-term sustainability goals.
This is where the fight becomes more than local. Energy policy is regional and national. Grid planning is multi-year and involves utilities, regulators, and sometimes federal agencies. Environmental standards are shaped by law and oversight mechanisms that extend beyond a single town. So even though the protests and hearings happen locally, the decisions are often influenced by broader policy frameworks.
That creates a strategic opportunity for opponents: they can push not only for project-specific changes but also for systemic reforms. For example, they may advocate for stronger requirements around energy sourcing and emissions reporting, or for clearer rules about how grid upgrades are funded and scheduled. They may push for moratoriums or stricter zoning criteria until capacity questions are resolved. They may also demand that companies provide more detailed information about long-term expansion plans rather than treating each phase as a separate surprise.
Proponents, for their part, may respond by emphasizing flexibility—promising to use renewable energy, to invest in grid upgrades, or to adopt cooling innovations. They may also argue that delays harm competitiveness and that digital infrastructure is essential for modern life.
But the tension remains because the planning system is often built for slower, more predictable development cycles. AI compute demand can be anything but predictable. It can surge based on product launches, model training trends, and shifting customer behavior. That unpredictability makes it harder to commit to long-term infrastructure decisions without creating risk for someone—either the community, the utility, or the company.
When the risk is unclear, trust becomes the battleground.
Trust is also why the “just beginning” framing matters. If the current resistance