The servers behind a cloud drive sit in a building that hums all night, keeps banks of diesel generators on standby, and draws water from the same aquifer as the houses next door. Virginia's legislative review found roughly one in three of the state's data centers sits near residential areas (Aethair). That proximity is the public health story.
A 2026 narrative review from University of North Texas researchers describes the core problem in plain terms: no published study has yet measured the low-frequency noise spectrum of an operating data center or examined health outcomes in the communities around one (Clean Technologies). The exposure is real and growing. The evidence base has not caught up, and neither have the noise ordinances, air permits, or water contracts that govern it.
This is not an argument against computing. It is an argument that a hyperscale campus is an industrial facility, and that industrial facilities next to residents carry obligations — acoustic, atmospheric, hydrological — that most U.S. jurisdictions have not yet written into law.
Why does a data center's hum reach so far into a neighborhood?
Standard A-weighted (dBA) measurement under-weights low frequencies, which is why resident complaints can run far ahead of what an official sound reading shows (KnowYourH2O). The practical reach is what makes the exposure a neighborhood issue rather than a property-line one: traffic noise may fade by 300 feet, while a data center hum travels 800 feet or more, with the greatest area of impact at 3,000 feet — about half a mile (St. Louis Public Radio).
What the body does with that exposure is better established than what data centers specifically do. A 2016 systematic review of observational studies found associations between low-frequency noise and self-reported annoyance, sleep problems, concentration difficulty, and headache in adults living near relevant sources, with pooled high annoyance around 10% of those exposed (KnowYourH2O). The Environmental Health Project's 2026 review links chronic noise exposure to tinnitus, sleep disturbance, elevated blood pressure, mental fatigue, cardiovascular disorders, and cognitive impairment (Environmental Health Project).
Sleep is where the mechanism becomes concrete. Noise that fragments sleep without waking the sleeper still suppresses the restorative stages, and repeated nights of it push blood pressure and stress hormones upward. Kasey Fowler-Finn, a Saint Louis University professor who studies low-frequency sound, put the physiological chain directly: such noise increases cortisol and cardiovascular risk and reduces cognitive functioning, especially in children (St. Louis Public Radio). She also measured the reach — traffic noise may fade by 300 feet, while a data center hum travels 800 feet or more, with the greatest area of impact at 3,000 feet.
The honest caveat matters for credibility. Research on pure infrasound at typical community levels is mixed — controlled trials often find minimal effects, while epidemiological work finds annoyance and sleep disturbance more consistently than chronic disease (KnowYourH2O). That ambiguity is precisely why the 2026 review calls for acoustic design standards rather than waiting on perfect evidence (Clean Technologies).
What comes out of the diesel generators?
Diesel backup generators emit nitrogen oxides (NOx), fine particulate matter (PM2.5), carbon monoxide, and diesel particulate matter when they test and run. Units can produce 200 to 600 times more NOx than a natural gas plant, and they run on schedules that neighbors do not control (Clarity). PM2.5 penetrates deep into lung tissue and crosses into the bloodstream, driving asthma exacerbations, heart attacks, stroke, and premature death; NOx is an ozone precursor, so it also degrades air quality downwind of the generation site.
The cumulative arithmetic is what turns a permit technicality into a public health question. Individual facilities often sit below pollution release limits, but clusters of them can exceed the emissions of a single gas-fired power plant (Environmental Health Project). One analysis found that data center backup generators in Virginia could already cause 14,000 asthma-related cases every year, with a public health burden estimated at $220–300 million annually (Clarity). Nationally, pollution from U.S. data centers could contribute to roughly 1,300 premature deaths a year by 2030 (Environmental Health Project).
There is a countervailing number, and it deserves airtime. Virginia's legislative review found generator emissions account for less than 4% of nitrogen oxides in Northern Virginia, since the units mostly run for scheduled testing (Aethair). That finding is contested: 2026 research from Virginia Commonwealth University found generator emissions in parts of the region rival power plant emissions, and state regulators concluded generators run more often than old permitting assumptions allowed. Regional averages hide fenceline exposures, which is the case for continuous local monitoring rather than annual permit totals.
How much water does a data center take from the local supply?
Evaporative cooling consumes water that is never returned to the source — it leaves as vapor. A single hyperscale data center can use 200 million gallons of water per year, and up to 57% of the cooling water at some facilities comes from potable sources, the same treated supply households drink and bathe in (Environmental Health Project).
The health consequence is indirect but not abstract. Water pulled into cooling towers is water not available for drinking, sanitation, and firefighting during a drought, and data center consumption patterns have been linked to increased waterborne disease risk, worse drought stress, and poor maternal and infant health outcomes (Environmental Health Project). The mechanism runs through municipal capacity: when a utility's treated supply is committed to industrial cooling, the margin left for household use and for sanitation during a shortage narrows, and sanitation failures are the pathway by which water stress becomes waterborne disease.
Disclosure is the missing link. Fewer than one-third of data center operators measured or reported their water consumption in 2021 (Environmental Health Project). A community cannot weigh a health tradeoff it cannot see, which is why water-use reporting requirements have become a standard ask in siting negotiations. The footprint also reaches beyond the fence line: facilities consume water to cool equipment and, indirectly, to generate the electricity that runs them, so a campus in a water-scarce region draws on two stressed systems at once (World Resources Institute).
Why do current regulations miss all three exposures?
Most noise ordinances are written in dBA, which systematically under-weights the low frequencies data centers produce, so a compliant facility can still generate the hum residents report (KnowYourH2O). Virginia's legislative review found the constant low-frequency noise is a genuine challenge for nearby residents yet typically does not violate existing ordinances, which were written for ordinary intermittent noise — and some localities are unsure they even have authority to regulate it through zoning (Aethair). Air permits carry a similar scale problem: a permit evaluates one facility against a threshold, while the health effect depends on what the cluster adds up to.
Regulation that reviews each site in isolation cannot see the cumulative total. A permit issued to a single facility can be fully compliant while the airshed it shares with a dozen other campuses is not — which is why the relevant question for a planning commission is the sum across all permitted generators, not any one application.
Exposure | How the harm reaches residents | Where regulation falls short |
|---|---|---|
Low-frequency noise | Continuous hum penetrates walls and fragments sleep, raising blood pressure and stress hormones over months of exposure | Ordinances measure in dBA, which under-registers the low-frequency band residents actually feel |
Diesel generator emissions | PM2.5 and NOx from routine testing reach the fenceline and form ozone downwind of the site | Permits evaluate one facility against a threshold rather than a cluster's cumulative load |
Cooling water | Potable supply diverted to evaporation reduces what remains for drinking and sanitation during drought | Water use often goes unreported, and contracts rarely weigh public health tradeoffs |
What should residents and local officials ask for?
Three asks convert the evidence into leverage: weighted noise measurement, cumulative emissions review, and mandatory water disclosure. None requires new science — only that a permit condition match the exposure it is supposed to control.
On noise, request that compliance be measured in dBC or narrowband analysis rather than dBA alone, with limits applied at the nearest residence rather than the property line, since low frequencies pass straight through conventional buffers (KnowYourH2O). The 2026 acoustic review argues the same point from the research side: standards should be set to protect health, using instrumentation that can actually detect what the facility emits (Clean Technologies).
On air, the question to put to a planning commission is cumulative, not individual: what do the generators across all permitted facilities add up to at the nearest school, playground, or senior community? Virginia's 2026 permitting revision, which requires new generators to meet limits at least as stringent as EPA Tier 4 standards, shows that generator controls are available now rather than aspirational (Aethair).
On water, ask for reported withdrawal volumes and the drought contingency plan in writing before approval, not after. Operators' own reporting rates make this a reasonable demand rather than a hostile one — most were not measuring consumption at all as recently as 2021 (Environmental Health Project).
The wider lesson is about sequencing. Community benefit agreements and modernized zoning codes work best before a campus is built, when a locality still has leverage over siting, noise enclosure design, and water contracts (World Resources Institute). After the ribbon cutting, the same conversation becomes a variance request, and the health evidence — however strong — arrives too late to shape the building.
1Why does a data center pass a noise test while neighbors still hear it?
An A-weighted meter is designed to approximate how the human ear perceives mid-range sound, so it deliberately discounts the very-low-frequency band that cooling fans and chillers produce. A facility can read compliant on dBA while residents still feel and hear a persistent hum. C-weighted or narrowband measurement captures that energy, which is why acoustic researchers recommend it for data center compliance testing.
2How can several compliant data centers produce an air quality problem together?
Air permits are typically issued facility by facility against a fixed threshold. The health risk depends on the sum of emissions across every permitted facility in the same airshed plus the grid power plants serving them. In dense clusters, the total can exceed what a single large plant would produce, even when every individual permit is technically satisfied.
3Is there a way to cool a data center without competing for drinking water?
Air-cooled designs use far less water than evaporative cooling but consume more electricity and generate more fan noise — a direct tradeoff between the water and acoustic footprints. Some operators use closed-loop or non-potable sources to reduce competition with drinking supply. Which design a facility chooses is usually decided at the permitting stage, which is why it belongs in the siting negotiation.
No comments yet. Be the first to share your thoughts!