The document that would reshape a slice of Lancaster, Pennsylvania, for a generation arrived in the hands of the people most affected by it roughly forty-eight hours before it mattered. That, at least, is what the local organising group Lancaster Stands Up said happened in November 2025: the community benefits agreement negotiated between the City of Lancaster and three data centre developers landed in front of residents only two days before the city council was scheduled to vote on it. Two days to read it, two days to understand it, two days to decide whether the numbers on the page — some twenty million dollars and change, spread across a community foundation and a clean-energy fund — amounted to a fair price for whatever the city was agreeing to host.
There is something almost clarifying about that detail. A community benefits agreement is, in theory, the mechanism by which a community says yes on its own terms. It is meant to be the opposite of imposition, the negotiated alternative to the bulldozer. And yet here was a yes being assembled in a room most residents never entered, printed, and handed to them at the last possible moment before it hardened into a binding contract. The people whose electricity bills and water table and night-time quiet were the subject of the negotiation were, in the most literal sense, presented with a fait accompli and asked to applaud.
I keep returning to those two days because they contain, in miniature, the question that the entire American build-out of artificial-intelligence infrastructure is now forcing communities to answer. The question is not really whether a data centre is good or bad, though that argument rages in county halls from Maryland to Texas. The deeper question is procedural and philosophical at once: when a corporation offers a town money in exchange for accepting a facility that will draw down its power, its water, and its air, who exactly has the standing to accept? And once a figure has been named — once there is a price at which the yes can be purchased — has the community consented to the data centre, or has it merely been compensated for one? Those are not the same thing. The gap between them is where this essay lives.
An Opposition That Doubled
To understand why the money has started flowing, you have to understand the wall the money is trying to get over. American communities have turned against the data centre with a speed that has startled the industry building them.
The numbers are stark and recent. According to Data Center Watch, a tracker maintained by the AI research firm 10a Labs, the count of active grassroots opposition groups across the United States more than doubled in a single year — rising from 396 at the end of 2025 to 833 by March of 2026, a surge that now touches forty-nine of the fifty states. In the first quarter of 2026 alone, that opposition blocked or delayed at least seventy-five projects worth a combined 130 billion dollars. The states with the densest concentrations of resistance, the researchers found, were Maryland, Ohio and Texas. As Fortune and NBC News both reported on the underlying data, this is not a spasm but a structural shift: communities have internalised an opposition playbook, and the setbacks logged in the first three months of 2026 already exceeded the whole of the previous year.
What makes the revolt so difficult for developers to dismiss is that it does not sit neatly on one side of the political map. As the environmental outlet Grist has documented, the backlash against data centres is emphatically bipartisan — a coalition of rural conservatives worried about land and water and progressive ratepayer advocates worried about bills, united by a shared suspicion that the benefits of the AI boom accrue elsewhere while the costs settle locally. In Texas, officials in Brazoria County unanimously rejected tax abatements for a proposed three-billion-dollar facility after residents and local leaders raised objections. In Virginia, the historic epicentre of the American data centre, public support has weakened sharply as residents contend with utility costs, infrastructure strain and a growing sense that they have lost control over their own landscape.
Opposition, in other words, has become a genuine business risk, and a large one. A project blocked at the county level is capital stranded. And so the industry has reached for an old tool from the world of stadiums, pipelines and wind farms: if a community can say no, perhaps it can be persuaded, in writing and for a fee, to say yes.
The Ledger Beneath the Handshake
The instrument for that persuasion is the community benefits agreement — a legally binding contract between a developer and a host municipality, or a coalition of local groups, that promises defined benefits in exchange for the project going ahead. These are not new. Columbia University's Sabin Center for Climate Change Law maintains a searchable database of publicly available community benefits agreements, drawn from across the American infrastructure landscape: offshore wind and solar, energy storage and transmission, waste facilities, transport, retail, and now, increasingly, data centres. The database exists precisely because these agreements are usually private, negotiated out of public view, and difficult to compare. Gathering them in one place makes the going rate visible.
And the going rate can be very large. One agreement on file with the Sabin Center commits a developer to paying a host community 169.9 million dollars over twenty-five years, including twenty-eight million dollars routed through a separate payment-in-lieu-of-taxes arrangement, alongside earmarks such as five million dollars for a park, ten million for a training centre, and five million for a university research partnership. That particular agreement, it should be said plainly, is not a data centre deal — it is the host community agreement signed in March 2023 between the Town of Brookhaven, New York, and the developer of the 924-megawatt Sunrise Wind offshore project. But the reason it belongs in this story is that the Sabin Center itself points to agreements like it as the template from which data centre deals are now being drawn. As the Center's Vincent Nolette wrote in a May 2026 analysis of community benefits agreements and data centre development, the practice of negotiating such agreements for data centres specifically is still emerging, with few direct precedents — so developers and communities are reaching into the archive of wind farms and waste plants and adapting the provisions they find there. The 169.9-million-dollar figure is not the price of a data centre. It is the shape of the ledger that data centre negotiations are being poured into.
The data centre deals themselves are already appearing, and Lancaster is the instructive case. There, as Nolette documented, the agreement bundled together 20.25 million dollars in commitments: ten million dollars plus a separate quarter-million contribution to the Lancaster County Community Foundation, and ten million to the city's own Sustainable Development and Clean Energy Fund. Reading across the Sabin database, the menu of what these agreements can contain is remarkably consistent regardless of the industry: financial contributions, which are by far the most common; workforce provisions requiring developers to hire local workers, raise wages, or contract with minority-owned firms; property-value guarantees for surrounding homeowners; and environmental controls setting limits on noise, air quality, water consumption and energy use.
Laid out like that, a community benefits agreement looks like exactly what its name promises — a community, benefiting. But notice what the structure quietly accomplishes. It takes a sprawling, open-ended, decades-long relationship between a town and an industrial facility, and it converts that relationship into a fixed and finite number. Twenty-five years, 169.9 million dollars. Everything the community might have felt, feared, or fought about is resolved into a figure and a term. That conversion is the whole point. And it is also where the trouble begins.
What Is Actually Being Priced
It would be a mistake — a category error, really — to read these payments as goodwill. A benefits agreement is not a gift. It is compensation, and compensation implies a harm being compensated for. To understand what a community is being paid to accept, you have to look at what a modern AI data centre actually does to the place that hosts it.
Start with electricity, because that is where the arithmetic is most brutal. The hyperscale facilities built for AI training and inference consume power on the scale of a mid-sized city, and in the mid-Atlantic grid operated by PJM Interconnection, that demand has detonated wholesale prices. An analysis by the Union of Concerned Scientists found that data centres were responsible for 63 per cent of the price increase in PJM's 2025/2026 capacity auction — translating into some 9.3 billion dollars in higher electricity costs borne by ratepayers across the region in a single year. Capacity prices, as the Institute for Energy Economics and Financial Analysis and reporting by E&E News have tracked, rocketed from 28.92 dollars per megawatt-day in 2024/25 to 329.17 dollars in 2026/27 — roughly a tenfold jump, and a 76 per cent year-on-year surge in the first quarter of 2026 alone. Absent intervention, the projections suggest PJM consumers could pay an additional 100 billion dollars through 2033, with a typical family paying something like seventy dollars a month more on their power bill by 2028. The Union of Concerned Scientists further found that upward of 95 per cent of the data centre projects it examined passed all of their transmission-connection costs straight onto local ratepayers.
Then there is water. A single large data centre can consume up to five million gallons a day for cooling — the daily draw of a town of fifty thousand people — and, by widely cited estimates, roughly two-thirds of data centres built since 2022 sit in water-stressed regions, where that draw is not an abstraction but a competition with farms and households for a shrinking resource.
Then there is the air. A study by researchers at Caltech and the University of California, Riverside, released on the arXiv preprint server, estimated that the air pollution associated with the electricity and backup generation feeding US data centres could cause as many as 1,300 premature deaths a year by 2030. A separate arXiv study by Guidi and colleagues, cataloguing the environmental burden of 2,132 American data centres, found the sector already accounts for more than four per cent of US electricity consumption — 56 per cent of it from fossil fuels — and generates over 105 million tons of carbon-dioxide equivalent, a carbon intensity 48 per cent above the national average. Fortune, reporting on a National Bureau of Economic Research working paper, put the annual hidden health and environmental damage from data centres at roughly 25 billion dollars, of which some 3.7 billion is attributable specifically to AI workloads.
This is the crucial point, and it is one that recent economic scholarship has begun to formalise. In their 2026 paper “Taxing Artificial Intelligence,” the researchers Juliette Faivre and Sarah H. Cen argue that AI's harms — among them the environmental pressures it places on local communities — represent real, unevenly borne costs, and they explore taxation as a mechanism for redistributing those costs back toward the people who shoulder them. That framing matters enormously for how we read a community benefits agreement. The money at stake is not a bribe for goodwill. It is a payment against a bill that is already being run up: in higher electricity rates, in water drawn from a stressed aquifer, in particulate matter in the lungs of the people downwind. The developer's cheque is, in the most precise sense, compensation for a harm that the negotiation has already priced.
Which brings us, at last, to the question that no dollar figure can settle.
Locke's Highway
If someone pays you for a harm they have caused you, have you consented to the harm? The intuition most of us carry says no. Compensation is what you receive after the fact, often through a court, for a wrong you did not agree to. Consent is what you give before, freely, that makes the act permissible in the first place. The two belong to different moral universes. And yet the community benefits agreement is built to make them look like one and the same — to fuse the payment and the permission into a single signature.
The philosophical tradition that underwrites the modern idea of consent runs through John Locke, and Locke saw the difficulty coming three centuries ago. In the Second Treatise of Government, Locke faced an awkward problem: governments claim authority over people who never actually signed anything. His solution was the doctrine of tacit consent. Express consent, Locke argued, makes a person a full member of a commonwealth — but even those who never expressly agree give their tacit consent simply by enjoying the benefits of a territory. By owning land, by inheriting property, indeed, in his famous formulation, merely by travelling freely on the highway, a person tacitly consents to the government whose roads they use and whose protection they enjoy.
Tacit consent is a brilliant and slippery idea, and it maps onto the data centre dilemma with uncomfortable precision. The developer and the local officials who sign a benefits agreement are, in effect, invoking a Lockean logic. The residents live here; they use the roads and the grid and the public services that the tax revenue funds; they enjoy the benefits the agreement provides; therefore they have tacitly consented to the arrangement their representatives negotiated on their behalf. The signature of a handful of elected officials stands in for the will of thousands who never entered the room.
But Locke's critics have always pressed on exactly the weak point that Lancaster's two-day warning exposes. Tacit consent, the philosopher David Hume objected in the eighteenth century, is a fiction when the alternative to accepting it is intolerable — a man who cannot leave the ship he was carried onto in his sleep has not consented to the captain's authority merely by remaining aboard. If consent is to mean anything, there has to be a meaningful capacity to withhold it. And that is precisely what the resident handed a finished contract two days before the vote does not have. She cannot renegotiate the terms. She cannot, realistically, sell her house and move because a data centre is coming. Her tacit consent, if we insist on calling it that, is manufactured by the very structure that claims to rest upon it.
The Coasean Bargain and Its Silence
If Locke supplies the political vocabulary, it is the economist Ronald Coase who supplies the machinery. The community benefits agreement is, at bottom, a Coasean bargain, and understanding it as one reveals both its elegance and its blind spot.
Coase's famous theorem holds that when property rights are clearly defined and the costs of bargaining are low, private parties can resolve an externality — pollution, noise, a factory's smoke drifting over a neighbour's field — through negotiation, and will arrive at an efficient outcome regardless of who initially holds the right. If a data centre imposes costs on a town, the Coasean solution is not to ban the data centre or to regulate it from above, but to let the parties bargain: the developer compensates the community for the harm, the community accepts the facility, and everyone is, in the aggregate, better off. The benefits agreement is Coase made flesh. The 169.9-million-dollar figure is the price at which the externality clears.
The trouble is hidden in Coase's own fine print, in the two conditions his theorem requires: clearly defined property rights, and low transaction costs. Both collapse when the “party” on one side of the table is not a person but a community. Who, exactly, holds the property right to a town's clean air, its aquifer, its night-time quiet, its degree of local control over its own future? These are not owned by anyone in particular; they are held in common, which is to say they are held by everyone and therefore, for the purposes of a negotiation, by no one in particular with standing to sell. The local official who signs the agreement is treated as the holder of a right that in truth belongs diffusely to thousands of residents, unborn children among them, who cannot sit at the table. And the transaction costs of actually assembling all of those people to bargain in their own name are, of course, precisely what make the shortcut of representation necessary — and precisely what the two-day warning economises upon.
So the Coasean bargain gets struck, but it is struck by a party that is standing in for the real owners of the thing being sold. This is the sleight of hand at the heart of the mechanism. The agreement converts a community's environmental and economic exposure into a fixed, time-limited financial transaction, and it does so through the signatures of a small number of officials on behalf of residents who were never party to the terms. The efficiency of the bargain depends on treating the community as a single unified will. Its legitimacy depends on the opposite being true — on those residents actually having authorised the deal. You cannot have both, and the agreement quietly banks on no one noticing which one it has sacrificed.
What Money Corrupts
Suppose we set aside the problem of standing. Suppose the residents did vote, in a fair referendum, with full information and ample time, and suppose they said yes to 169.9 million dollars. Would the philosophical difficulty dissolve? A second tradition of thought says no — that there are some goods whose nature is changed, even damaged, by the act of putting them up for sale, and that consent purchased is not always consent at all.
The most prominent contemporary voice here is the Harvard political philosopher Michael Sandel, whose 2012 book “What Money Can't Buy” mounts a sustained argument against what he calls the drift from a market economy to a market society — a society in which market values crowd out the non-market norms that govern civic life. Sandel's central worry is that commodification can be corrupting: attaching a price to certain goods does not merely allocate them, it alters our relationship to them. His most quoted example is a study of Israeli day-care centres that introduced a fine for parents who collected their children late. The fine was meant to reduce lateness. Instead lateness increased — because the fine had reconfigured a moral obligation into a mere price. Parents who had once felt guilty for keeping a teacher waiting now felt they had simply purchased extra time, fair and square. The norm had been crowded out by the number, and, tellingly, when the fine was later removed the lateness did not revert: the moral relationship, once monetised, did not grow back.
Apply Sandel's logic to a benefits agreement and something unsettling comes into view. When a town's relationship to its own air and water and self-determination is expressed as a price, that relationship may be permanently transformed. The question “should we allow this facility that will strain our aquifer and raise our neighbours' bills?” is a civic and moral question, the kind a community deliberates. The question “will we accept 169.9 million dollars for it?” is a transaction. Reframing the first as the second does not merely answer the question; it changes what kind of question it is. And once a community has learned that its environmental burden has a market price, the moral weight that might have led it to refuse — the sense that some things about a place are not for sale — may not survive the lesson.
The philosopher Elizabeth Anderson sharpened this intuition into a theory decades earlier. In her 1993 book “Value in Ethics and Economics,” Anderson argued that we value different goods in different ways, not merely in different amounts, and that the market's great flattening — its insistence on reducing every value to a single quantity, a price — fails to do justice to the plurality of ways that things can matter. Some goods, she argued, are properly treated as commodities; others, when we treat them as commodities, are degraded by the treatment. Her examples ran to commercial surrogacy and to the cost-benefit analysis of environmental protection — cases in which reducing a good to its market price expresses the wrong kind of valuation and damages the good in the process. A community's control over its own environment and its collective future arguably belongs on Anderson's list of goods that resist commodification without loss. To ask “how much?” is already to have decided that the answer is a number, and to have foreclosed the possibility that the honest answer is “not at any price.”
None of this means the money is worthless or that communities are wrong to take it. A town facing a data centre it cannot stop may be entirely rational, even wise, to extract the largest possible payment for a harm it will suffer regardless. The point is subtler and more corrosive. It is that the payment, by resolving the matter, obscures the fact that a genuine act of collective consent may never have occurred at all — and that the smoothness of the transaction is precisely what allows everyone to pretend that it did.
The Officials Who Hold the Pen
Everything, in the end, comes back to the signature, and to the small number of hands that hold the pen.
The structural reality of the benefits agreement is that it is negotiated by a handful of local officials — a mayor, a few council members, a county board — acting as agents for a principal, the community, that is too large and too internally divided to negotiate for itself. This is the ordinary logic of representative government, and there is nothing inherently sinister about it. We elect people precisely so that they can make decisions we cannot all make together. But representation carries a demanding condition: the represented must be able to hold their representatives to account, which requires, at minimum, that they know what is being decided in their name and have a real opportunity to object before it is final. The two-day warning in Lancaster is the sound of that condition being quietly waived.
There is a deeper asymmetry, too, one that no amount of procedure fully cures. The officials who sign are, for the most part, present-day adults who will benefit soonest from the money — the park that opens next year, the training fund that pays out this decade, the property-tax relief that shows up on the next bill. The costs, meanwhile, are back-loaded and diffuse: the aquifer drawn down over twenty-five years, the grid infrastructure whose bills arrive gradually, the children who will inherit both the facility and the exhausted terms of the agreement long after the cheques have cleared. A twenty-five-year contract signed by people who will not personally bear most of its later years is a bargain struck across a boundary — between those who consent and profit now and those who will live with the consequences later — that no signature can legitimately cross. Locke's tacit consent at least imagined a living person walking a living highway. The benefits agreement asks the unborn to have consented too, by the mere fact of being born here.
This is why the growth of opposition from 396 groups to 833 in a single year should be read not simply as nimbyism, the reflexive refusal of change, but as something closer to a jurisdictional revolt — a mass assertion that the people signing do not, in fact, speak for the people affected. When residents in forty-nine states organise to block facilities their officials were prepared to welcome, they are making a claim about standing. They are saying that the pen was held by the wrong hands, or held on the wrong terms, or moved too fast across the page. The benefits agreement was supposed to be the sophisticated, grown-up alternative to that revolt — the mechanism that would convert conflict into contract. The revolt's persistence suggests that a great many people can tell the difference between having been consulted and having been paid.
When Everything Has a Number
Return, one last time, to those two days in Lancaster. What the compressed timeline reveals is not merely bad process, the kind of thing a good-government reformer might fix with a mandatory ninety-day comment period and a public hearing. It reveals the fault line running beneath the entire enterprise of buying a community's yes.
A community benefits agreement makes a genuine promise and performs a genuine service. It gets money to places that will bear real costs, and it does so more reliably than the alternative, which is often nothing at all. If a data centre is coming regardless — and against the tide of capital now flooding into AI infrastructure, many of them are — then a town is plainly better off with a hundred million dollars than without it. Nothing in political philosophy obliges a community to refuse compensation for a harm it cannot prevent. To pretend otherwise would be its own kind of cruelty, a purism paid for by the people least able to afford it.
But we should be precise, even ruthless, about what the money does and does not accomplish, because the industry has a strong interest in blurring exactly this line. The payment compensates. It does not, by itself, consent. It answers the question of how the costs will be distributed without ever honestly asking the prior question of whether the costs should be incurred, or by whom that decision was legitimately made. It treats a community as a single owner of goods — clean air, water, quiet, self-determination — that no single party actually owns, and it treats the signatures of a few present officials as the settled will of a multitude that includes people who never spoke and people not yet born. And in expressing all of it as a number, it risks doing what Sandel warned of: changing a civic question about what a place is willing to become into a transaction about what a place is willing to accept, and eroding, in the process, the very capacity for refusal that would make a real yes meaningful.
The existence of a price at which acceptance can be purchased does change the nature of consent in this context, and it changes it in a specific and troubling direction. It relocates the decision from the many to the few, from the future to the present, from the register of deliberation to the register of exchange. It offers the form of agreement while quietly dispensing with its substance. A community that has been paid has not necessarily agreed; it has been settled with. And the more fluently our institutions learn to price a town's environmental exposure — the more polished the benefits agreement becomes, the larger and more generous its figures grow — the easier it becomes to mistake the settlement for the agreement, and to forget that there was ever a difference.
That difference is worth defending, precisely because it is so easy to sell. The cost of yes is not the number on the contract. The cost of yes is what we agree to stop asking once the number has been named.
