Seventy percent of Americans say they don't want a data centre near them, and a major reason is the constant low-frequency hum from cooling systems that neighbours report as sleepless nights, headaches, and nausea.

The rise of AI has, over the past three years, produced a physical footprint most people did not particularly see coming. Every large language model that answers a question, every image generator that produces a picture, and every recommendation algorithm that decides what appears in front of you online runs on hardware that sits in actual physical buildings somewhere. Those buildings are called data centres. And Americans, according to recent polling, have quite strong feelings about having one built anywhere near their own house.

According to a Gallup poll of 1,000 US adults conducted between 2 and 18 March 2026, and published on Gallup’s own website in May of that year, seventy-one per cent of Americans oppose the construction of an AI data centre in their local area. Forty-eight per cent oppose it strongly. Only twenty-seven per cent support one being built nearby. The opposition holds across almost every demographic slice Gallup measured. It holds across age, race, education, income, and urbanicity. It cuts across party lines, with Democrats slightly more opposed than Republicans, but neither party close to a majority in favour. In the same poll, respondents were also asked about a nuclear power plant being built nearby. Fifty-three per cent opposed a local nuclear plant. Which means more Americans, according to the current data, would rather have nuclear infrastructure built in their neighbourhood than a data centre.

The reasons cluster around a small set of concerns. Half of opponents cite excessive resource use, with eighteen per cent mentioning water and eighteen per cent mentioning electricity. Sixteen per cent cite pollution, including noise. About one in five point to quality-of-life concerns. And a similar share cite economic effects, meaning higher utility bills or the tax breaks these facilities have negotiated from local governments.

The noise, on closer analysis, is the concern most people have never seen coming.

What the neighbours are actually reporting

According to an analysis by the Environmental and Energy Study Institute, a Washington-based nonprofit that provides briefings to members of Congress on energy and environmental policy, data centres emit a kind of sound that ordinary regulatory instruments have significant difficulty measuring. The noise comes from three main sources. Large industrial cooling systems and air chillers, which run continuously to prevent the servers from overheating. Whirring cooling fans, arrayed in banks that can number in the hundreds. And, at facilities powered by their own on-site generators, either diesel back-ups or increasingly natural-gas turbines that some observers have compared to jet engines.

The combined output can reach ninety-six decibels at the property line. That level runs twenty-four hours a day, seven days a week. It does not stop for the night. It does not fluctuate with the working day. It is, in the residents’ descriptions, closer to an aircraft holding a landing pattern permanently overhead than to any ordinary industrial noise their neighbourhoods had previously encountered.

What EESI’s analysis documents is that residents living near these facilities have been reporting a consistent pattern of complaints. Headaches. Sleep disturbance. Ear pain. Elevated blood pressure. Nausea and vertigo in a smaller number of cases. In Chandler, Arizona, residents of the Brittany Heights subdivision tried noise-cancelling headphones and earplugs, and eventually blocked a second proposed facility from being built nearby. In the Great Oak subdivision near Manassas, Virginia, residents downwind of an Amazon Web Services complex have documented four hundred and twenty-four separate air exhaust vents on the roofs of the facilities generating the hum they hear at night.

It’s not only the hum that affects the environment – it’s the water consumption. Check out this video we made on the topic:

The most-cited example in the popular coverage of this issue is not, strictly speaking, an AI data centre. In Granbury, Texas, a Bitcoin cryptocurrency mining facility running roughly sixty thousand computers less than one hundred yards from a mobile-home park produced complaints of vertigo, nausea, high blood pressure, migraines, and fluid coming from the ears sufficient to prompt a lawsuit filed in 2024, which is still ongoing. Cryptocurrency mining facilities and AI data centres are legally and technically distinct, but they use similar large-scale cooling infrastructure and produce similar noise profiles. Which is why the Granbury complaints keep appearing in coverage of the AI data centre issue, and why residents worried about AI infrastructure keep pointing to Granbury as an example of what they fear happening in their own neighbourhoods.

The difficulty for residents pursuing complaints through ordinary noise ordinances is that most such ordinances were written for a different kind of problem. They regulate maximum peak decibel levels and typically assume the noise source will fall silent at night. A data centre does not fall silent. It is designed to run at constant load indefinitely. Which means the ordinance limit, applied to any single moment, is technically met, while the cumulative exposure builds up over months and years in a way the regulatory framework was never designed to capture.

What the noise research actually establishes

The health-effects picture is more nuanced than the popular coverage suggests, and worth handling carefully. Some of the reported symptoms sit on very solid medical ground. Some of them do not.

The World Health Organization’s environmental-noise guidelines, published in 2018 for the European region and widely referenced elsewhere, identify chronic exposure to environmental noise above roughly fifty-three decibels at night as a well-documented cause of sleep disturbance, cognitive impairment, and increased long-term cardiovascular risk. The mechanism is not mysterious. Chronic noise keeps the sympathetic nervous system in a state of low-level activation. Cortisol runs slightly elevated. Deep sleep and REM sleep are interrupted by micro-arousals the sleeper does not consciously register. The cardiovascular system adjusts, over months and years, in a direction that raises the long-term risk of hypertension and heart disease. All of this applies to ordinary audible noise, and it applies to residents living near a data centre for exactly the same reasons it applies to residents living near a motorway or an airport.

A subset of the complaints, however, concerns something else. Sound below roughly twenty hertz is called infrasound. It sits beneath the threshold of ordinary human hearing. Some residents living near data centres and similar facilities have reported physiological symptoms they attribute to infrasound exposure, including nausea, dizziness, and a felt vibration in the chest. The evidence base for infrasound-specific health effects at the levels typically emitted by industrial cooling systems is, on the current published research, contested. Some studies find effects. Some do not. Regulatory bodies including the WHO have declined to set specific exposure guidelines for infrasound because the evidence has not yet consolidated. Which does not mean the residents are wrong about what they are experiencing. It means the specific causal mechanism they are attributing it to has not yet been definitively established.

What the evidence more clearly supports is the broader claim that chronic exposure to twenty-four-hour audible industrial noise produces measurable health effects that ordinary residential neighbourhoods were not built to tolerate. According to a 2025 cross-sectional study by Lasin Faizal, Anil Kumar Rama Krishnappa, Medha Chakraborty and Shanmukha Koppolu, published in Bioinformation and hosted on the National Institutes of Health’s PubMed Central repository, titled “Impact of Urban Noise Pollution on Sleep Disorders and Neurological Symptoms”, urban residents exposed to noise levels above sixty-five decibels show consistently higher rates of sleep disturbance, headaches, dizziness, and irritability than lower-exposure controls. The study is not specific to data centres, but the noise-exposure levels its participants were experiencing sit in the same range as the residential-boundary levels documented near several major data-centre installations.

The political consequence has moved faster than the industry expected. About a dozen US states have introduced moratoriums or restrictions on large-scale data centre development, including Maine, Oklahoma, and New York, whose governor Kathy Hochul in 2026 imposed a one-year pause on new large-scale permits. Local jurisdictions have imposed their own moratoriums. According to Data Center Watch, an initiative from AI security firm 10a Labs that tracks community opposition to data centre projects, roughly ninety-eight billion US dollars in proposed projects across twenty individual developments were blocked or delayed in the second quarter of 2025 alone. That single-quarter figure exceeded the cumulative total of all opposition-driven delays and cancellations Data Center Watch had recorded across the previous two years combined.

None of which is going to slow the underlying demand. The AI infrastructure the country is building continues to need physical buildings full of servers, and those buildings continue to need to be somewhere. What the polling and the noise complaints together suggest is that the industry has, without particularly noticing, entered a new phase of its expansion. The technical work remains what it was. The social work has shifted. The question is no longer whether the machines can be built. It is whether the people who will end up living next to them will let them be built.