While the internet is being flooded with bland and soulless AI-generated text, out in the real world people of flesh and blood are mobilizing against the very infrastructure on which all that slop depends. Because out there — offline, outdoors — AI generates ever more noise and friction, right in its physical “home”: the data center. This is where AI becomes visible, audible, tangible.
At a town hall meeting in the US state of Ohio, where angry residents were protesting the construction of a new data center, one speaker recently cut to the heart of the matter: “A big employer who uses the water of 50,000 people… which only hires about ten people is not an employer… They are an extraction.”
The protest is by no means confined to a single location. An interactive map recently launched by the legendary environmental activist Erin Brockovich documents thousands of complaints from US citizens about existing or proposed data center sites. And in Europe, Asia and South America, too, resistance is mounting.
Because the ratio of water consumption to jobs is just one of many troubling symptoms that data centers impose on communities. Beyond the strain on local water reserves, data centers also mean:
- Strain on the power grid — their enormous electricity demand can push regional grids to their limits, even to the point of blackouts
- Strain household budgets in the form of rising energy prices, driven by the sharp surge in demand and the grid investments it forces
- On top of that: light pollution, soil degradation and noise, as fans and generators produce a constant humming and whirring
- And health risks: on-site fossil power generation (e.g. diesel generators), releases particulate matter and other air pollutants that promote respiratory and cardiovascular disease
The list of “external effects,” to borrow a euphemism from economics, is impressive. It takes considerable imagination to make a credible case that data centers benefit a community. Nor does the argument hold that data centers in economically weak locations make substantial contributions to municipal budgets through taxes. Because taxes are neither a substitute for jobs nor compensation for emissions.
And so far we’ve only been talking about the visible symptoms. Yet the problem runs deeper. It raises a fundamental question: what legitimizes the existence of these data centers? Do they deserve their “license to operate”?
My argument is that data centers are part of a business model that, at every stage, extracts more than it creates. And what does society get back? Spoiler alert: for the time being, remarkably little.
The most tangible link in a hungry business model
At first glance, one might dismiss the protests against data centers as just another facet of the NIMBY phenomenon — “not in my backyard,” the opportunistic resistance of residents to infrastructure projects that surfaces only when those projects are built in their own surroundings. And indeed, in the US, property owners face partial expropriation or even the demolition of their homes to make way for new high-voltage power lines – lines that are meant primarily to feed the energy hunger of the data centers.
But this is about far more than individual self-interest. Data centers raise fundamental questions about their value to the community — at every step of the value chain in which they are embedded. That they draw the public’s ire is owed to the fact that they are the most tangible link in that chain.
The chain begins earlier
Extraction doesn’t begin with the building of roads and the excavation of earth for new data centers, and it certainly doesn’t end there.
The first grab for resources happened during the training of the data models on which today’s generative AI rests — the very AI that is driving the data center boom. Abuse took place at this stage already; whether merely in a moral sense or also a legal one is something the courts are currently sorting out. What is beyond dispute is that the data models were fed with millions of works that the tech firms used without asking their creators for permission, let alone paying them compensation.
In a technologically impressive tour de force, the models chewed their way through thousands of pages of texts carefully conceived and formulated by human minds, swallowed hard-won research findings, devoured countless hours of video, spent weeks dissecting songs into sound bites, and tore millions of images into pixels.
Today the devoured material is recombined statistically — but of course without any mention of its origin — and spat back out in the answers to our prompts.
A threat to some, a promise to others
The masterminds behind this technology boast that they will render superfluous not only the very people whose works they used without authorization. Or, to put it in the words of Mira Murati (ex-OpenAI): eliminate jobs that perhaps should never have existed in the first place.
No. If Dario Amodei, CEO of Anthropic, is to be believed, AI is expected to displace 50% of entry-level jobs within the next five years. Similar threats have come from Sam Altman of OpenAI.
But what is a threat to workers is a promise to investors. The people whose
livelihoods are under threat are not the target audience; they are just the collateral damage of these aggressive forecasts. According to independent economists such as Nobel laureate Daron Acemoglu, these scenarios are unrealistic. Yet some of the loudest “job-apocalypticians” are preparing their companies for IPOs and are desperately hunting for capital. By presenting “number of jobs eliminated” as a yardstick for efficiency, they hope to lure in investors.
Meanwhile, by the way, both Amodei and Altman are now walking it back. Perhaps they’ve realized that, for weighty institutional investors like pension funds, it isn’t quite so attractive after all to invest the money of insured workers in companies that boast of abolishing them.
But what about the AI that cures cancer?
Now one might ask: perhaps the data center in my community does contribute to a “greater whole” — to the common good. Perhaps the very high-voltage line I don’t want on my property delivers the energy for that much-cited “AI that cures cancer.” Or for the AI that will “eradicate poverty.” Or for AI’s groundbreaking contribution to “solving climate change.”
Don’t worry: the chance that your protest against a data center will prevent the cure for cancer is slim.
Because the data centers putting a strain on infrastructure serve primarily generative AI. It is what drives the building boom; this is widely documented. And this kind of AI is especially energy-intensive. At the same time, it is a “general purpose” technology: usable for almost anything, but not strictly necessary for solving real problems.
For real problems, single-purpose AI is often enough. Long before the tech industry became obsessed with developing resource-intensive models, specialized systems were rendering important services: cancer detection, prediction of extreme weather events, acceleration of drug development, to name just a few. And these systems have little in common with the gigantic data centers being protested today — neither in terms of resource consumption nor in terms of purpose.
As a rule, also claims about AI’s positive climate effects should be taken with a grain of salt. Most of them refer to precisely these older, specialized systems — and not to the generative AI that today accounts for the lion’s share of energy consumption.
In any specific case, the sobering answer to the question of what’s running inside the data center is: we don’t know. Because from the outside, there’s no way to assess what kind of AI runs in a given data center. Residents don’t know what kind of AI they’re sacrificing their land for, which AI they’re sharing their power grid and water reserves with, or what purpose the high-voltage line in their garden serves. Because the industry doesn’t tell them. Yet transparency about what goes on inside data centers would be a minimum requirement to earn goodwill among the local population. And by minimum I mean necessary, but by no means sufficient.
As long as not even these questions are settled, the highly creative comparisons with which AI evangelists try to play down the massive resource consumption are of little use. When Sam Altman says that the energy use of a ChatGPT query amounts to one second of running an oven, and others stress that data centers’ water consumption is tiny compared to agriculture, the only question that arises is: who gets fed by tokens?
And so the circle closes
This is the question that gets ignored at schools. While the parents at the town hall meeting loudly protest the construction of data centers, AI integrates itself seamlessly and frictionlessly into the laptops in their children’s schools. Unobtrusively, AI offers its help the moment they start typing. This is made possible by the contracts that educational institutions have signed with Big Tech. Mastering AI, so the saying goes, is crucial for the workforce of the future: “AI won’t replace you — someone who’s better at AI than you will.” That threat does its work.
But only until graduation day, when they’re served up a commencement speaker from the orbit of Silicon Valley. One from exactly that guard which, elsewhere, boasts of how many jobs its technology renders superfluous. Before the graduates he chooses gentler words, to be sure: he speaks of industrial revolution and of progress with no alternative.
But they don’t buy it. Because they know that these people keep working toward having AI take over their jobs — the same AI built on the works of their ancestors, used without their consent, and the same AI to which their parents’ property is meant to be sacrificed. No wonder they drown him out with boos.
Rightly so.