Communities Resist Forced Datacentres Over Energy Strain
This matters because datacentres are critical for AI and economic growth but are opposed by local communities who are affected by their strain on energy resources and climate impact.
We are told that the datacentre is now the fixed price of admission to the age of artificial intelligence - that Scotland’s glens and India’s cities must yield their power lines and their water tables to it, or forfeit the wealth the technology promises, as if compute could be conjured without concrete and copper. But notice the instrument that has actually arrived this summer, in the fields of Scotland and the districts of India where communities have taken to the streets against facilities they say were forced upon them: it is not merely a building, it is a purpose-built apparatus for training and running models at a scale no laboratory, no university cluster, no prior arrangement of transistors could sustain. That is a genuine capability, and it should be named honestly before it is defended or attacked. The question worth asking is not whether the datacentre is holy or wicked but what, precisely, it does that could not be done five years ago, and what that doing costs in the specific grid, in the specific summer, in the specific village whose taps run dry when the cooling towers spin up.
The story celebrates that the datacentre was built - the concrete poured, the racks installed, the power substation connected, the ribbon somewhere cut before a minister and a regional director from some cloud company. But a made thing does not stop where its maker’s attention stops; it goes on drawing current, drawing water, drawing heat off its own machinery long after the launch photograph is filed. The question the celebration skips is the only one that lasts: who is answerable for what this facility does to the town around it, and what did its makers fail to imagine when they modelled it only as capacity, never as neighbour?
This summer’s disputes in Scotland and India benefit, by the industry’s own reckoning, an enormous population: every person who will eventually use an AI system, plus the shareholders of the big tech companies building the infrastructure, plus governments hoping to attach their names to growth figures. Set against this vast diffuse constituency stands a much smaller, sharply defined one: the residents of the villages and towns where the machines actually sit, who lose water, grid capacity, quiet, and in some cases the land itself. The arithmetic looks lopsided in favour of the many until you ask the four questions the calculus insists on asking of any pleasure or pain: how intense, how certain, how near, and how long.
The story frames the datacentre boom as natural consolidation - the necessary infrastructure of a maturing digital age, no more sinister than the laying of railway track. But look at what is actually being fenced: the electricity grid, the water table, the shared atmospheric budget that villages in Scotland and towns across India have depended upon and helped steward for generations, now claimed as feedstock for machines few of those villages will ever operate. The question the framing skips is who benefits from the fence - and whose access it removes.
The situation is described as a growth story interrupted by unreasonable neighbours. Beneath the description, two forces are in tension: the appetite of big tech for computation without limit, and the finitude of water, grid capacity, and the patience of people who live beside the substations. The equilibrium between them is the actual state of affairs, and the stability of the datacentre boom is what this tension looks like to those who are not standing outside the fence in Scotland or in a village in India watching the transformer hum through the night.
The official account, offered with confidence by the companies concerned and echoed by governments this summer from Edinburgh to Delhi, is that datacentres are simply infrastructure - pipes and wires for a digital economy that will make everyone richer, the way railways once did, or the electric grid before them. Communities in Scotland and India who object are, on this telling, understandably nervous but ultimately misinformed; they will thank the industry later, once the AI dividend arrives.
The story arrives as a fight between AI’s economic promise and the communities standing in the way of it - Scotland and India both site datacentre protests as the local cost of a global technology gold rush, with governments caught between hyperscaler investment and the households whose taps and lights depend on the same grid. That is the framing, and it is not wrong exactly. It is just built to keep one detail at the edge, where it does not have to answer for anything.
Jeremy Bentham
This policy benefits [the hyperscaler and its shareholders] by [the value of accelerated computation and reduced latency]. It harms [the community solar co-op in Scotland] and [the residents of rural India] by [deferred renewable integration and potential curtailment of essential cooling during peak heat]. The arithmetic is uncomfortable, but the arithmetic is the argument.
The opponent, speaking through the House, presents a structural observation: that the grid connection queue is not a new invention of AI, but an existing bottleneck into which new, powerful applicants are inserted, altering the ranking. This is the strongest point made. It is empirically true that the queue existed before the large language model. It is also true that a government relations team outranks a parish newsletter. To deny this is to deny the physics of bureaucracy. I concede that the mechanism of priority allocation is not inherently novel; it is ancient. The steam engine demanded coal; the telegraph demanded wire. The novelty lies not in the queue, but in the velocity and scale of the demand that now floods it.
However, the House stops at the description of the mechanism. They describe the how - the queue, the control room, the shift schedule - but they hesitate at the why. They note that state electricity boards make decisions under stress without an “AI ethics framework.” This is a complaint about the absence of a framework, not an analysis of the welfare outcomes of the decision itself. Here, our frameworks diverge sharply. The House seems to imply that the lack of a specific, named ethical rubric for “AI” is a defect in governance. I argue that the defect is not the absence of the word “AI,” but the absence of the Calulus.
When the control room in India faces peak heat, the operator does not ask, “Is this fair to the metaphorical concept of Artificial Intelligence?” They ask, “Whose feeder gets curtailed?” The House treats this as a moral vacuum. I treat it as a clear moral calculation. The pain of a household without cooling during a heatwave is intense, immediate, and physical. The pleasure of a hyperscaler’s reduced latency is diffuse, delayed, and largely economic. The intensity of the bodily suffering of the poor, who lack air conditioning and whose labor depends on the heat not killing them, outweighs the marginal efficiency gain of the machine. The operator’s decision, made in a control room with a shift schedule, is actually the most honest moment in the entire process: it is a naked aggregation of suffering. It is not “unethical” because it lacks a framework; it is potentially unjust if the aggregation is wrong.
The House suggests that the problem is the “ranking” altered by the hyperscaler. But let us count the people. In Scotland, the community solar co-op represents a collective of citizens attempting to secure energy independence. Their delay is a pain of frustration and lost revenue. The hyperscaler represents capital seeking profit. If the grid cannot carry both, and the queue favors the one with the strongest voice, we have a failure of representation, not just a failure of physics. The pain of the co-op members is real; the pleasure of the hyperscaler’s shareholder is real. But the co-op members are likely many; the shareholder is one. The principle of “each person counts for one” demands we weigh the co-op’s aggregate happiness against the shareholder’s. The House observes the disparity in power but offers no metric to resolve it. I offer the metric: the intensity and duration of the pain.
Consider the panopticon. I designed it not to punish, but to observe. The grid queue is a panopticon of demand. We must make the suffering visible. Currently, the suffering of the rural Indian household during a blackout is hidden in the shift schedule. The suffering of the Scottish co-op is hidden in the waiting list. If we were to mandate that every curtailment decision be accompanied by a public accounting of who suffered and who benefited, the “ranking” would change. The government relations team would find that their arguments carry less weight when the aggregate suffering of the populace is laid bare.
The House warns that we cannot write an ethics framework for every technological shift. I agree. We do not need an “AI Ethics Framework.” We need a “Greatest Happiness Framework.” The latter is universal; it applies to the steam engine, the telegraph, and the transformer. The former is a distraction, a label that allows us to pretend the problem is new when it is merely old. The pain of the heatwave is the same pain it was in 1920. The only difference is that the machine demanding the cooling is now smarter, not more moral.
Therefore, the legislator’s task is not to create a new category for AI. It is to enforce the existing law of utility upon the new technology. Prioritise the feeder that serves the greatest number of human beings over the feeder that serves the machine that serves the few. If the grid is too small, build it. But do not let the queue become a mechanism for the rich to buy the air of the poor. The arithmetic is clear: the many must outweigh the one, regardless of how loud the one can shout into the control room.
house-style
The ethicist frames the datacentre dispute as a calculus of diffuse benefit against concentrated harm, arguing that the immediate, certain pain to local communities - water, grid capacity, quiet - must outweigh the future, probabilistic gains of growth and infrastructure. One notices the ethicist’s insistence on the “four questions” of intensity, certainty, nearness, and duration. With that detail load-bearing, the argument reads not as a moral inquiry but as an accounting exercise that ignores the ledger’s missing column: the opacity of the system itself.
The ethicist is correct that the harms to the immediate community are immediate, certain, and continuous. The village in India watching its water table drop is not suffering a theoretical risk; it is suffering a physical depletion that begins the day the machines switch on. This is a concrete detail, a load-bearing fact: the local infrastructure is being leveraged for a global computation without a corresponding local return, creating a fiscal and physical deficit that the “growth figures” cannot fill. The ethicist’s arithmetic is sound regarding the tangible loss. Where the framework diverges is in the treatment of the “benefits.” The ethicist treats the benefit to the “enormous population” as a valid counterweight to the local harm, assuming a net positive utility calculation is possible. This is the first error. The second error is the assumption that the system’s operation is transparent enough to be weighed against community stability.
Consider the engineering principle of ballast as defined by Wigardus à Winschooten in 1681: stabilising weight in the hold, but also een onnutte Ballast - a useless burden to the world. The datacentre acts as ballast in the literal sense, drawing down local resources to stabilise the global computation. But when the metaphor decouples from the engineering function, the “benefit” becomes folk language. The community does not see the stabilising weight; it sees the burden. The ethicist’s calculus assumes that the diffuse benefits are real and measurable. They are not. They are projections, contingent on a future adoption curve that has not yet been proven. The certainty of the local harm is high; the certainty of the global benefit is low. To treat them as commensurable is to ignore the asymmetry of evidence.
The deeper divergence lies in the concept of scale. The ethicist implies that the scale of the benefit (the entire population) justifies the scale of the harm (the local community). This is a category error. Scale in engineering is not a moral multiplier; it is a stress factor. A system that scales by externalising its costs onto a single node is not scaling; it is failing to encapsulate. The datacentre is not a node in a distributed network that shares load; it is a sink that absorbs local resources without returning value. The ethicist’s framework, by focusing on the distribution of pain and pleasure, misses the structural integrity of the system. A system that cannot hold the cost of its own operation within its boundaries is not a system; it is a parasite.
There is a Dutch phrase, schaap met vijf poten, for the candidate who has every necessary quality. The phrase exists because the candidate doesn’t. The datacentre, in the ethicist’s vision, is such a candidate: beneficial to all, burdensome to none. But the reality is that the burden is borne by the few, and the benefit is claimed by the many. This is not a failure of intent; it is a failure of design. The design document did not say how the local costs would be internalised. It said what the system would do: compute. It did not say how it would sustain itself. On a system of that size, that omission is the same as saying it would not.
The plain question is this: if the local community bears the entire burden of the system’s operation, and the global population bears no cost, who is the stakeholder? If the stakeholder is defined by who bears the risk, then the ethicist’s framework has identified the victim but failed to identify the responsible party. The answer is not found in the distribution of benefits, but in the allocation of liability. Until the liability is internalised, the “benefit” is a fiction, and the “harm” is the only real thing.
The people in the room - the engineers, the planners, the ethicists - are executing a system that is wasting them. They believe they are building the future. They are not. They are extracting the present. The warmth for these people is real; they are not clowns, but they are operating in a room that has been designed to hide the load-bearing details. The detail here is not the water table or the electricity bill. The detail is the lack of a contract. There is no contract between the datacentre and the village that says the village will be compensated for the loss of its future. Without that contract, the system is not an engine of progress; it is a machine of extraction.
The ethicist’s framework is useful for identifying the harm. It is useless for preventing it. To prevent the harm, one must move beyond the calculus of pain and pleasure and into the architecture of liability. The system must be designed to hold its own costs. Until then, the “benefits” are just noise, and the “harms” are the signal.
Transmission note: The distinction between scale as a moral argument and scale as a stress factor is not merely semantic. It is the difference between a system that survives and a system that collapses. Read the design document. Look for the liability clause. If it is not there, the system is not ready.
The Verdict
Where They Agree
Both Bentham and The House concur that the localised harms inflicted by datacentres - on water tables, grid stability, and community infrastructure - are immediate, certain, and concrete. They share the view that these costs are not speculative future risks but present-day realities that begin the moment the facilities are switched on. This shared premise is significant because it forces the debate away from abstract technological promise and grounds it in observable, material consequences.
both debaters operate from an unstated assumption that the current system is fundamentally opaque. Bentham’s call for a public accounting of every curtailment decision and The House’s demand to know “which named project… loses its place” are different expressions of the same diagnosis: the mechanisms for allocating resources and distributing pain are hidden from public view and democratic accountability. They agree that power operates in this obscurity, but they propose radically different tools to illuminate it - Bentham’s utilitarian calculus versus The House’s engineering audit of liability.
Where They Fundamentally Disagree
The nature of a valid counterweight to local harm. The empirical disagreement is whether the purported global benefits of AI-driven datacentres are real, measurable, and certain enough to be placed on a scale against local costs. Bentham treats them as potentially real but discounted by uncertainty and diffusion, making them a weak counterweight. The House treats them as fictional projections, “contingent on a future adoption curve that has not yet been proven,” and thus not commensurable with the harm at all. The normative disagreement is deeper: Bentham believes a valid utilitarian calculation is both possible and necessary, where the pleasure of the many can, in principle, outweigh the pain of the few if the arithmetic supports it. The House rejects this entire framework, arguing that a system that externalises its costs is structurally unsound and immoral by design, regardless of the arithmetic; the problem is the lack of a liability contract, not an imbalance in utility.
The primary tool for achieving justice. The empirical element here concerns what is most effective at changing outcomes: better information or better design. Bentham is empirically committed to the idea that transparently publishing the “aggregate suffering” caused by allocation decisions would shame the system into a more equitable ranking. The House is empirically skeptical, implying the system’s design is so entrenched that only a fundamental redesign - hardwiring liability and cost-internalisation into the “design document” - can work. Normatively, they champion different master principles. Bentham’s highest value is aggregate welfare, measured by his utilitarian calculus. The House’s highest value is systemic integrity, where a system’s moral worth is judged by its ability to contain its own operational costs without parasitically draining its host.
Hidden Assumptions
- Jeremy Bentham: Assumes that a transparent public accounting of costs and benefits would lead powerholders to make more utilitarian decisions. This is contestable; if public scrutiny does not alter the underlying power dynamics or economic incentives, the result could be a performative disclosure that changes nothing.
- Jeremy Bentham: Assumes that the “greatest happiness” is a neutral, universal framework that can be cleanly applied without its own political commitments. This is contestable; the act of quantifying and comparing disparate forms of “happiness” and “pain” is itself a value-laden exercise that can mask bias.
- house-style: Assumes that a system can be designed to fully “hold its own costs,” making liability and benefit perfectly contiguous. This is contestable; all large-scale infrastructure, from railways to power grids, involves some degree of cost externalisation and diffuse benefit, suggesting the ideal of a perfectly encapsulated system may be an impossibility.
- house-style: Assumes that the “diffuse benefits” of AI are largely fictional and are merely “noise.” This is contestable; if the economic and productivity gains from AI infrastructure are real and significant for a large number of people, then dismissing them entirely is its own form of myopia.
Confidence vs Evidence
No confidence-evidence mismatches were flagged. Either both debaters calibrated their claims carefully, or neither used explicit confidence markers - making every claim equally weighted, which is itself a form of overconfidence.
What This Means For You
When reading about conflicts over datacentres, your first question should not be about the technology itself, but about the contract: what is the specific mechanism for compensating the host community for its lost water and strained grid, and is that compensation delivered upfront or promised for later? Be highly suspicious of any coverage that treats “economic growth” as an unexamined, monolithic good that automatically offsets local harm. The most important data point to look for is the one everyone is hiding: the name of the specific energy project that was delayed or cancelled to fast-track the datacentre’s grid connection.