The cloud has always been a beautiful lie. Not a malicious one, just a convenient metaphor that let us stop thinking about what's underneath. Somewhere along the way, we collectively agreed that computing happens "in the cloud," a weightless, invisible utility like air conditioning or Wi-Fi. You don't think about where it comes from. You just use it. But the cloud is not weightless. It is concrete, steel, water, and diesel generators. It hums. It heats up. It drinks from the same aquifers your neighborhood does. And now, with more than 4,000 AI data centers already operating across the United States and another 3,000 planned or under construction, the fiction is getting harder to maintain. Communities are pushing back, and they're winning.

The backlash nobody predicted

For years, data center development followed a predictable formula: secure land, negotiate power, build. Community engagement was barely an afterthought. These were "clean" facilities, not factories with smokestacks. They brought tax revenue, required few permanent employees, and didn't clog local roads with truck traffic. What's not to like? Quite a lot, it turns out. In Q2 2025 alone, grassroots opposition blocked or delayed an estimated $98 billion worth of proposed data center projects across the United States, according to Data Center Watch. By the end of the year, at least 48 projects worth roughly $156 billion had been cancelled or placed on hold. That's up from just six cancellations in 2024 and two in 2023. The opposition is not fringe. A November 2025 Morning Consult survey found that 41% of American voters favored a ban on AI data centers near their homes, up from 37% just one month earlier. Wall Street has noticed too. A March 2026 New York Times investigation noted that the top AI companies have forecast spending $710 billion on data centers across North America in 2026, but "rising skepticism from neighbors" could put a hitch in those plans. This is no longer a local zoning dispute. It's becoming a political movement.

What communities are actually worried about

The concerns are concrete, not abstract. They cluster around three things: power, water, and quality of life. Power. A single modern AI data center can consume as much electricity as 100,000 homes. US data center energy use hit approximately 176 terawatt-hours in 2023, about 4.4% of total national electricity consumption, and projections suggest it could double or triple by 2028, reaching up to 12% of the country's electricity use. That surge doesn't just stress the grid. It raises electricity prices for everyone connected to it. As investor Chamath Palihapitiya put it bluntly: local residents are being asked to subsidize AI infrastructure through higher electricity bills with no upside. Water. Data centers need enormous volumes of water for cooling. A medium-sized facility can consume up to 110 million gallons per year, equivalent to roughly 1,000 households. In Newton County, Georgia, a single Meta data center uses 500,000 gallons per day, which accounts for 10% of the entire county's water consumption. The county continues to receive permit requests for facilities that would more than double its total water use. Quality of life. These are industrial-scale facilities, sometimes spanning dozens of acres, running 24/7 with backup diesel generators, transformer hum, and bright security lighting. In Virginia, the epicenter of US data center growth, activists have described them as "a plague" on neighborhoods. In California, a report found that 82% of operational and planned data centers sit in communities already burdened with poor air quality, and 65% are in areas facing the highest groundwater threat levels. The economic benefits that were supposed to offset these costs have often been overstated. A 2025 report from the University of Michigan's Science, Technology, and Public Policy program found that data centers typically generate few permanent jobs relative to the scale of public subsidy they receive. The Brookings Institution echoed this, noting that in many communities, "the overriding concerns seem to be more about the loss of a community's way of life and local control over its future."

The irony of "cloud computing"

There's a deep irony here that's worth sitting with. The entire branding of modern computing was designed to make infrastructure invisible. "The cloud" evokes something ethereal, something that exists everywhere and nowhere. It was a masterstroke of marketing. But the infrastructure was never invisible to the people who live next to it. It was only invisible to the people who use it from a distance. The metaphor worked precisely because it created a gap between consumption and consequence. That gap is closing. When your electricity bill doubles because a hyperscaler moved in next door, the cloud stops being abstract very quickly.

Jevons paradox, in concrete

This is also a vivid, physical illustration of the Jevons paradox playing out in real time. In 1865, economist William Stanley Jevons observed that improvements in the efficiency of coal-powered steam engines didn't reduce coal consumption. Instead, by making coal cheaper and more productive, efficiency drove demand up. The same dynamic is now unfolding with AI compute. When DeepSeek released its R1 model in early 2025, claiming performance comparable to leading models at a fraction of the training cost, the initial market reaction was panic: Nvidia lost $600 billion in market value in a single day. The logic seemed straightforward, more efficient AI means less demand for chips and data centers. But that's not what happened. Cheaper and faster models lowered the barrier to entry, enabling more companies to train AI, more teams to experiment, and more products to incorporate AI features. Total demand for compute grew, and with it, the physical infrastructure required to support it. As a Nature Cities paper published in 2025 documented, algorithmic efficiency gains in data centers may enlarge, not shrink, the overall energy footprint of AI. Efficiency doesn't shrink the footprint. It expands the frontier. And that frontier is being built in someone's backyard.

The Singapore problem

Not every country has the luxury of sprawl. Singapore, one of the most digitally connected nations on earth, imposed a moratorium on new data center builds back in 2019. The reasoning was simple: on a land-scarce island with finite energy and water resources, you cannot just keep building. The moratorium was partially lifted in 2022, with the government allocating a carefully rationed 80 megawatts of new capacity through a structured application process. In late 2025, Singapore opened applications for an additional 200 megawatts, but with a significant catch: at least 50% of proposed capacity must be powered by eligible green energy pathways. Singapore's approach is instructive because it treats data center capacity as a strategic resource to be allocated, not a market outcome to be accepted. It's a sharp contrast to the American model, where the default has been to build first and manage consequences later. But Singapore's constraints also highlight a geopolitical dimension. In a world where AI capability depends on physical infrastructure, the countries that can build data centers fastest have a strategic advantage. Data sovereignty, the principle that a nation's data should be stored and processed within its borders, increasingly depends on having the physical capacity to do so. For small, land-scarce nations, this creates a genuine tension between digital ambition and physical reality. The result is a dual-location strategy emerging across Southeast Asia: low-latency inference workloads in Singapore, high-latency training workloads in neighboring Johor, Malaysia, where energy and space constraints are less severe. The cloud, it turns out, has a geography.

The fracking parallel

The closest historical analogue might be hydraulic fracturing. A decade ago, fracking was the technology that was going to deliver American energy independence. The economics were compelling. The geopolitical benefits were real. And then communities started experiencing the externalities: contaminated groundwater, earthquakes, noise, plummeting property values. What began as scattered local opposition coalesced into a national political issue that reshaped energy policy. Data centers are following a remarkably similar trajectory. The technology delivers genuine economic value. The national strategic interest is real. But the costs are concentrated in specific communities while the benefits are diffused across the entire digital economy. That asymmetry is the engine of political backlash. The opposition is also notably bipartisan. A Fortune investigation found that NIMBY revolts are turning voters against data centers in Republican strongholds, not just progressive enclaves. When your water table drops and your electricity bill spikes, partisan affiliation becomes secondary.

What comes next

None of this means data centers will stop being built. The economic and strategic imperatives are too strong. But the era of frictionless expansion is over. What's likely to emerge is a more contested, negotiated process. Communities will demand more: guaranteed rate caps on electricity, water usage limits, local hiring commitments, and meaningful revenue sharing rather than token tax incentives. Some jurisdictions will impose moratoria, as Singapore did. Others will extract concessions that fundamentally change the economics of data center development. The tech industry will need to get much better at community engagement, something it has historically treated as a box-ticking exercise. The Brookings Institution recommends treating rural communities as "partners with meaningful input and leverage," not just sites to be acquired. And the rest of us might need to reckon with an uncomfortable truth: the services we use every day, the AI assistants, the streaming platforms, the cloud storage, have a physical cost that someone, somewhere, is bearing. The cloud was never weightless. We just weren't the ones carrying it.

References