AI Data Centers can be small, environmentally sustainable & community owned

The opposition is right about the harms. It is incomplete about the alternative — and the alternative is already being built.

Something is working. In the first quarter of 2026 alone, community opposition blocked or delayed at least 75 data center projects representing roughly $130 billion in planned investment — the highest quarterly total since tracking began in 2023. More than 300 cities, towns, and counties have enacted bans or moratoriums on hyperscale facilities. On July 14, New York became the first state to hit pause statewide, with Governor Hochul ordering a halt of up to one year on new facilities drawing 50 MW or more, pending environmental review.

Residents organized. They showed up at zoning hearings with utility bills and sound meters. They won.

We should say plainly: those wins are legitimate, and the grievances behind them are not imaginary. But a moratorium is a pause button, not a plan. And if the only thing our movement offers is no, we will win the hearings and lose the decade — because the compute is going to get built somewhere, by someone, on someone’s terms. The question is whether those terms are ours.

Part One: The Grievances Are Real

Let’s not soften this. The complaints against hyperscale campuses are grounded in measurable harm.

Electricity. U.S. data center power demand is climbing from roughly 31 GW in 2025 to 41 GW in 2026, pushing data centers’ share of peak summer demand from 4.1% to 5.3%. Gartner projects data center electricity consumption will grow 26% in 2026 alone. Meanwhile, average residential electricity prices are rising from about 17.29¢/kWh in 2025 to 18.02¢/kWh in 2026, and they are rising fastest precisely where large loads cluster. In PJM territory, utilities across seven states passed billions in transmission upgrade costs — incurred to connect data centers — onto the general ratepayer base. When your neighbor’s bill goes up to wire in a facility that employs thirty people, “externality” stops being an abstraction.

Water. Google’s Council Bluffs campus — one facility, in one Iowa county — has withdrawn an average of 3.9 million gallons of water a day and consumed 2.8 million of them. That consumption figure is the one that matters: roughly 80% of what an evaporatively cooled facility withdraws is lost to evaporation and never returns to the watershed. In a drought year, in a county negotiating irrigation allotments, a few million gallons a day is not a rounding error.

Noise. This is the harm that gets dismissed and shouldn’t be. Residents of Dowagiac, Michigan have taken a data center operator — Alliance Cloud Services, a subsidiary of a company named, without irony, Hyperscale Data — to federal court over a constant low hum that has run 24 hours a day for roughly two years; the city has fined the operator under its industrial noise ordinance, and the company has offered to buy neighbors’ homes. Microsoft faces a similar proposed class action in Wisconsin from residents living within 1.5 miles of a facility. Chronic exposure in the 50–70 dB range is associated with sleep disruption, headaches, vertigo, and hypertension. You cannot close a window against a low-frequency drone.

Anyone who tells a community these concerns are NIMBYism has not read the filings.

Part Two: But “No” Has a Ceiling

Here is what a moratorium does not do.

It does not lower anyone’s electric bill. It does not produce a single kilowatt of clean generation. It does not create a job, fund a library, heat a building, or put an asset on a community’s balance sheet. It moves the project one county over, to a jurisdiction with less capacity to negotiate — often a poorer one. And it leaves the underlying demand entirely intact, because the demand is not coming from the data center. It is coming from us.

There is a harder truth inside this. Every organization now fighting a hyperscale campus is, almost without exception, running its email, its case management, its payroll, and increasingly its writing and analysis on infrastructure exactly like the one it is opposing. The opposition and the demand are the same people. That is not hypocrisy — it’s the ordinary condition of living inside a system you’d like to change. But it does mean that refusal alone cannot be the whole strategy. You cannot boycott your way out of infrastructure you depend on. You can only replace it.

Which brings us to the framing that has trapped this entire conversation.

Part Three: The Doom-or-Boom Trap

Public argument about AI has collapsed into two stories, and both of them are demobilizing.

Boom says the technology is inevitable, the buildout is progress, and any friction is Luddism. Its practical instruction to you is: get out of the way.

Doom says the technology is a catastrophe, the buildout is enclosure, and the only honest posture is refusal. Its practical instruction to you is: hold the line and wait — for a regulator, a court, a federal bill, a moratorium that becomes permanent.

Notice that both stories end with the public as spectators. In one you’re an obstacle; in the other you’re a petitioner. Neither hands you a hammer.

The Copernican move here is not discovering new facts. It’s moving the center. Stop asking whether AI infrastructure gets built and start asking who owns it, where it sits, what it’s cooled with, and where the surplus goes. Once you move that center, an enormous amount of impossible-seeming complexity becomes simple, because the design questions have known answers. We have built community-owned infrastructure before. Rural electric cooperatives wired the countryside that private utilities wouldn’t touch. Municipal water systems, public libraries, and community broadband networks all exist because someone decided a shared necessity shouldn’t be a private toll road.

Compute is now a shared necessity. Treat it like one.

Part Four: What the Other Kind of Data Center Looks Like

The alternative isn’t hypothetical, and it isn’t a scaled-down hyperscaler. It’s a different design object with different physics and different economics.

Small, and sited near the load. Instead of a 500-acre campus drawing a gigawatt in a rural county, a distributed fabric of facilities in the hundreds of kilowatts, located in the communities they serve — attached to libraries, community colleges, municipal buildings, affordable housing developments, industrial parks. Small enough to fit the existing grid interconnection. Small enough that no new transmission line gets socialized onto ratepayers. Worth noting: New York’s moratorium threshold is 50 MW. A community-scale node sits roughly two orders of magnitude below the line that triggered a statewide pause — which tells you the objection was never to computers.

Closed-loop or air-cooled, not evaporative. The water problem is largely a cooling-architecture problem. New builds are migrating quickly toward closed-loop liquid and air cooling because the political cost of evaporative consumption now exceeds its efficiency savings. At community scale, zero-water-consumption cooling is straightforwardly achievable.

Waste heat as a product, not a nuisance. This is the part that converts a liability into a civic asset, and it is the single most persuasive thing you can put in front of a skeptical council. Direct-to-chip liquid cooling produces outlet temperatures of 50–60°C — warm enough to feed low-temperature district heating. The precedents are operating right now:

• In the UK, an edge computing firm installs immersion-cooled micro data centers inside public leisure centres, coupled to the pool’s heat load. One Devon host pool cut its gas consumption by 62%. Octopus Energy has committed £200 million to extend the model to roughly 150 pools.
• Stockholm’s Open District Heating scheme recovers excess heat — data centers are its largest single source — sufficient to warm 30,000 apartments a year.
• In Finland, heat-pump plants capturing waste heat from a Microsoft campus now supply roughly 40% of annual heating demand for 250,000 people across three municipalities. Google’s Hamina facility supplies 80% of local district heating demand, free.
• Paris 2024 heated the Olympic Aquatic Centre in Saint-Denis, plus the surrounding district, with recovered data center heat — about 10,000 MWh a year, enough for more than 1,000 homes.

Regulators have begun to notice. Germany’s Energy Efficiency Act requires data centers above 300 kW of connected IT load to hit a minimum Energy Reuse Factor of 10% if they begin operating on or after July 1, 2026 — rising to 15% for facilities starting in 2027 and 20% in 2028. In the United States, Virginia’s HB323 is the first state legislation on data center heat reuse, directing the Department of Energy to convene a working group and deliver a heat-reuse strategy.

Renewable-anchored, and generously sized. Pair the facility with local solar, wind, and storage — and deliberately over-build the generation relative to the compute load. The compute becomes the anchor tenant that makes the microgrid financeable; the surplus power flows to households, not to a distant utility. An islandable microgrid also keeps clinics, schools, and emergency services running when the grid goes down.

Community-owned, with local governance. A cooperative, a municipal utility, a nonprofit authority, a public-private partnership with real public equity. The point isn’t the legal wrapper — it’s that the revenue recirculates and the operating decisions are made by people who live within earshot.

That last item is what separates this from corporate greenwashing. A hyperscaler can install heat recovery and still extract every dollar of value out of the county. Ownership is the variable that decides where the money lands. Energy, data, and community governance have to be designed to operate in harmony rather than in extraction — which is a design choice, not an accident.

Part Five: You Do Not Have to Wait for the Government

The reflex, once someone accepts that the alternative is real, is to ask which agency will fund it. That’s a reasonable question and a slow one. Federal policy will move when it moves. Meanwhile:

Nothing in law prevents a library system from hosting a compute node today. Nothing prevents a community college from teaching the deployment and maintenance curriculum that staffs it. Nothing prevents a credit union from underwriting a solar-plus-storage installation, a housing authority from putting racks in a mechanical room, a co-op from selling surplus power to its own members, or a church with a large roof from becoming a generation site. These are contracts and site surveys, not acts of Congress.

What has been missing is not permission. It is coordination — and coordination is a thing civil society already knows how to build.

Part Six: The Coalition Is the Technology

The social science here is unglamorous and well-tested. Complex problems that no single organization can solve get solved by what practitioners call collective impact, and the model has five conditions: a common agenda, shared measurement, mutually reinforcing activities, continuous communication, and a backbone organization that holds it together. Nonprofits bring community trust and standing. For-profits bring capital, engineering, and the discipline of having to make the numbers work. NGOs and foundations bring convening power, patient money, and legitimacy. None of the three can do this alone; all three together can move faster than a legislature.

Map that onto a mini data center and the roles fall out almost mechanically:

Partner	What they contribute
Community nonprofit / 501(c)(3)	Backbone coordination, listening sessions, grant eligibility, public trust
Library, school district, community college	Host sites, workforce pipeline, civic legitimacy, existing broadband
Local for-profit firms	Engineering, installation, O&M contracts, an actual business model
Housing authority / affordable housing developer	Sites, anchor demand, direct low- and moderate-income benefit
Credit union / CDFI / municipal utility	Financing, metering, interconnection, bill-credit mechanics
Foundations & NGOs	Patient capital, shared measurement, cross-site learning
Residents	Governance, consent, priorities — and the mandate that makes it legitimate

This is not theoretical for us. Occupy AI is this kind of federation — a working coalition of responsible-AI practitioners spanning Community Internet, Altgov.AI, CommunityAgents.AI, mySPARKlab.AI, the Raise the Standard podcast, and Townhall Community. Different legal forms, different revenue models, one common agenda. The infrastructure arm is Verdant Data, operated by the Broadband Institute Foundation, a 501(c)(3) educational nonprofit — and the design framework above is drawn from its Verdant Data Center Whitepaper.

And because someone always asks: yes, the for-profit participation is a feature, not a compromise. Volunteer enthusiasm does not survive contact with a five-year maintenance schedule. Something has to pay for the technician. We’ve argued the economics of this at length in The Counter-Stack and the workforce model in Teach, Earn, Own. The short version: open technology plus local ownership plus a real revenue model is a more durable structure than either charity or extraction, because every party has a reason to keep showing up.

Part Seven: What to Do on Monday

If you are currently organizing against a hyperscale project, keep going — and add a second half to your ask.

1. Change the demand from “stop” to “stop, and here is what we’ll host instead.” A council that has only heard opposition hears a cost. A council that has been handed an alternative hears a choice.
2. Inventory your assets. Who in your community has roof space, a mechanical room, fiber, a substation, an interconnection queue position, or a heating load? Pools, greenhouses, laundries, and older buildings on hot-water heat are the highest-value waste-heat partners.
3. Name the backbone. Somebody has to own the calendar and the shared spreadsheet. Usually a nonprofit. Do this before the enthusiasm peaks.
4. Get one anchor tenant to say yes in writing. A library, a college, a housing authority. One letter of intent unlocks conversations that a good idea cannot.
5. Insist on the boring conditions. Water consumption at zero. Sound limits at the property line, written into the agreement, with measurement. Heat reuse commitment. Local hiring. Community equity or revenue share. Get them in the document.
6. Convene the room. If you want help running that first meeting, Townhall Community will help you organize one.

The Invitation

The world doesn’t change one person at a time — it changes as networks of relationships form among people who discover they share a common cause and a vision of what’s possible.

The hyperscale buildout is betting that those networks won’t form fast enough. That opposition will stay local, stay reactive, and stay strictly negative — a series of individual towns saying no while the aggregate demand routes around them. That bet is reasonable, and it has been right before.

It doesn’t have to be right this time. The engineering is solved. The financing structures exist. The host sites are sitting in every town, mostly under-used. The workforce can be trained at a community college in a year. What’s missing is the decision to organize — nonprofit with for-profit with NGO with neighbor — around a common agenda instead of a common enemy.

Doom asks you to wait. Boom asks you to move aside. We’re asking for something harder and considerably more interesting: build the thing, own the thing, and let the results argue for themselves.

---

Occupy AI is a coalition of responsible-AI practitioners convened by the Broadband Institute Foundation (dba Community Internet), a 501(c)(3) educational nonprofit. We are non-partisan by design: affordable power, quiet neighborhoods, local jobs, and community-owned infrastructure are not red or blue issues. Infrastructure work runs through Verdant Data.

Sources

• Data Center Projects Face ‘Unprecedented Surge’ in Blocks and Delays — Newsweek
• Can Growing Community Backlash Quiet the AI Data Center Boom? — Data Center Knowledge
• Data center opposition gains momentum as power demand spikes — S&P Global
• From Energy Use to Air Quality, the Many Ways Data Centers Affect US Communities — World Resources Institute
• US Data Center Power Demand Projected to Double by 2027 — Goldman Sachs
• Data Center Electricity Consumption to Grow 26% in 2026 — Gartner
• Data Center Water Use — MOST Policy Initiative
• Datacenter Water Usage 2026 — Kova Stack
• Michigan residents sue AI data center emitting noise 24/7 — Tom’s Hardware
• Communities Are Raising Noise Pollution Concerns About Data Centers — EESI
• Thermal Energy Networks Turn Data Center Waste Heat into a Hot Commodity — EESI
• Data Center Waste Heat Recovery 2026: District Heating Economics — Energy Solutions Intelligence
• Maximizing waste heat recovery from a building-integrated edge data center — Scientific Reports
• Collective Impact — Stanford Social Innovation Review
• A Cross-Site Study of 25 Collective Impact Initiatives — ORS Impact