Is It Safe to Live Near a Data Center? What the Evidence Actually Shows

Start with the complaints themselves, because they are remarkably consistent across states, operators and project types. The pattern in the cases we track runs in roughly this order.

Noise, first and loudest. In Mount Pleasant, Wisconsin, three Sturtevant residents filed a class action over the sound from Microsoft's 1.2 million square foot Fairwater campus, seeking to represent every household within a 1.5 mile radius, more than 1,000 homes. Microsoft acknowledged that residents north of the facility had noticed a tonal humming sound and attributed it to cooling fans. See our case file on the Fairwater noise class action. In Southaven, Mississippi, a proposed class of more than 10,000 residents sued over the gas turbines powering xAI's Colossus 2, describing roaring, rumbling and low frequency vibration around the clock and reporting readings above 70 dBA at their property lines. That case is the Southaven turbine noise class action.

Water, second. Four families near Meta's Stanton Springs campus in Newton County, Georgia sued in late August 2026, alleging that clearing, blasting and excavation disturbed the aquifer and filled their wells with silt, sediment and clays. See the Newton County well lawsuit. In Virginia, a Louisa County homeowner sued Amazon in July 2026 over brown well water, dust and noise from campus construction, described in that case file. In Oregon, Amazon Data Services agreed to a $20.5 million settlement, denying wrongdoing, in a class action over nitrate in groundwater in the Lower Umatilla Basin, covered in our nitrate settlement file.

Dust, vibration and the rest. Construction phase complaints about blasting, truck traffic and airborne dust show up in nearly every one of these disputes, and then fade when the site is finished, which is one reason they are often dismissed by developers as temporary. They are also the phase in which wells fail.

Sorting this honestly requires three separate columns, not one.

Category	What is in it	How strong is the evidence
Documented and measurable	Chronic noise and low frequency sound at the property line, sleep disruption, combustion emissions during generator testing and turbine operation, construction dust and vibration, well interference and water quality complaints, property value effects where impacts are real, electricity cost shifts	Strong. These are recorded in permits, ordinances, sound readings, water tests, utility filings and sworn complaints.
Genuinely unresolved	Long term health outcomes attributable specifically to living near a data center	Absent. To our knowledge no peer reviewed epidemiological study exists.
Not supported	Radiation or EMF harm from the facility, and claims that a data center is inherently a toxic site	Contradicted by the physics and by federal health agencies.

The middle row is the one that gets misreported in both directions. A PubMed search run for this guide on September 3, 2026 for the terms data center, residents and health returned 189 records, and none of them studied the health of people living near a data center. That absence is not evidence of safety and it is not evidence of harm. It means the study has not been done. The reasons are ordinary: hyperscale campuses at this density are recent, exposures differ enormously from site to site, affected populations are small, and the pollutants and sound involved are not unique to data centers, so researchers studying them have studied them elsewhere.

What that leaves is a bank shot. We know a great deal about what environmental noise does to sleep, and about what nitrogen dioxide and fine particles do to lungs and hearts, from decades of research on traffic, power plants, wind turbines and industrial sites. Applying that knowledge to a data center neighbor is reasonable inference. It is not the same as a study of data center neighbors, and anyone who tells you otherwise, in either direction, is overreaching.

If there is one health-adjacent claim that rests on real science, it is this one. WHO's guidance for the European Region recommends less than 30 A-weighted decibels in bedrooms for quality sleep, and less than 40 dB(A) as an annual average night noise level, Lnight, outside bedrooms. WHO's 2009 Night Noise Guidelines for Europe set 40 dB Lnight outside as the target to protect the public including the most vulnerable groups, and recommend 55 dB only as an interim target for countries that cannot reach 40 in the short term. WHO identifies sleep disturbance, cardiovascular effects, reduced school and work performance, and hearing loss among the outcomes associated with environmental noise, and notes consistent evidence that noise impairs children's cognitive performance.

Two things follow from those numbers that people usually get backwards.

First, hearing damage is not the issue. NIDCD states that sounds at or below 70 dBA are unlikely to cause hearing loss even after long exposure, and that long or repeated exposure at or above 85 dBA can. NIOSH sets its occupational recommended exposure limit at 85 dBA averaged over an eight hour workday. Most data center noise complaints involve levels well below the hearing damage range. The documented harm sits far lower on the scale, in the range where sound does not injure the ear but does keep a person from sleeping.

Second, the baseline matters more than the absolute number. The National Park Service explains that a 10 dB increase represents a tenfold increase in sound level and roughly a doubling of perceived loudness, and it has measured ambient levels of about 20 dBA on remote trails in Grand Canyon National Park. A rural property whose nighttime background is close to that will experience a 50 dBA nighttime limit as an enormous change, even though 50 dBA would be unremarkable in a city. This is why rural communities and urban ones report the same facility so differently, and why a limit that looks generous on paper can still be the difference between sleeping and not.

The closest thing to a controlled study of neighbors of a humming industrial installation is Health Canada's Community Noise and Health Study of wind turbines, which surveyed 1,238 households in Ontario and Prince Edward Island with a 78.9 percent participation rate. Calculated outdoor wind turbine noise reached 46 dBA at wind speeds of 8 meters per second, with C-weighted levels ranging from 24 to 63 dBC. The findings are worth stating precisely, because both camps quote half of them. Health Canada found statistically significant exposure-response relationships between rising noise levels and high annoyance, including annoyance from noise, vibration, blinking lights and shadow. It did not find noise level associated with self reported sleep, use of sleep medication, diagnosed sleep disorders, self reported illnesses, perceived stress, quality of life, hair cortisol concentrations, blood pressure, resting heart rate, or measured sleep. Annoyance itself was statistically related to several self reported health effects including blood pressure, migraines, tinnitus and dizziness.

That is a nuanced result, and the nuance is the point. Sustained noise reliably makes people miserable, and misery is correlated with reported symptoms, but a large government study looking for a direct physiological signature of the noise itself did not find one. A data center is not a wind turbine, and the sound signature differs, so this is an analogy rather than an answer. It is the best analogy available.

One more structural fact: there is no federal noise standard protecting your bedroom. EPA's noise program lost its funding in 1982, though the Noise Control Act of 1972 and the Quiet Communities Act of 1978 were never repealed and remain on the books unfunded. That is why every limit that actually binds a data center is a county or municipal one, which we catalog in our guide to data center noise limits by jurisdiction, and why the same facility can be legal in one county and a violation a mile away. Our noise claim center covers the legal options.

Data centers themselves do not burn anything. Their engines do. Every campus has backup diesel generators that are tested on a schedule and run during outages, and a growing number of AI campuses now run on-site gas turbines as primary power. Those are combustion sources, and the health literature on what they emit is extensive.

EPA describes nitrogen dioxide as one of a group of highly reactive gases formed by burning fuel, including emissions from power plants and off-road equipment. On health, EPA states that breathing air with a high concentration of NO2 can irritate airways, and that short term exposures can aggravate respiratory diseases, particularly asthma, leading to coughing, wheezing or difficulty breathing, hospital admissions and emergency room visits. Longer exposures to elevated concentrations may contribute to the development of asthma and potentially increase susceptibility to respiratory infections. People with asthma, children and the elderly are at greater risk.

On particulate matter, EPA defines fine particles as those 2.5 micrometers in diameter or smaller, notes they can get deep into the lungs and some may reach the bloodstream, and links exposure to premature death in people with heart or lung disease, nonfatal heart attacks, irregular heartbeat, worsened asthma, decreased lung function and increased respiratory symptoms. The populations most at risk are people with heart or lung disease, children, older adults, minority populations and low socioeconomic status populations.

The federal ambient standards give you a yardstick for what a regulator considers acceptable in outdoor air:

Pollutant	Primary standard	Averaging time
PM2.5	9.0 micrograms per cubic meter	1 year, annual mean averaged over 3 years
PM2.5	35 micrograms per cubic meter	24 hours, 98th percentile averaged over 3 years
PM10	150 micrograms per cubic meter	24 hours
Nitrogen dioxide	100 parts per billion	1 hour, 98th percentile of daily maximum, 3 year average
Nitrogen dioxide	53 parts per billion	1 year, annual mean

Here is the honest caveat, and it is a large one. Every health finding above comes from research on these pollutants in general, from traffic, power generation, industry and off-road engines. None of it comes from measuring the air around a data center and following the neighbors. Nobody has published that study. What can be said with confidence is that generator and turbine fleets emit pollutants that are known to be harmful at sufficient concentration, that the concentration at any given house depends on stack height, distance, weather, fleet size and how often the engines run, and that those variables are knowable from permits rather than guessable. Our guide to data center diesel generators covers the fleet numbers, the emissions tiers and how to pull the air permit for a facility near you, and our air pollution claim center covers the legal theories.

Water is where the documented harm gets most concrete, because water can be tested and a test result is a number on a page.

Two mechanisms show up. The first is physical disturbance during construction. Clearing, blasting and excavation on a large site can disturb an aquifer and push silt, sediment, clays and fines into nearby private wells, which is the allegation in both the Newton County, Georgia case against Meta and the Louisa County, Virginia case against Amazon. The second is contamination from operations or disposal, which is what the Oregon nitrate class action alleged about wastewater routed to Port of Morrow disposal systems.

Nitrate is worth understanding because it is the contaminant with the clearest federal number attached. EPA's maximum contaminant level for nitrate in drinking water is 10 milligrams per liter, and EPA states that infants below six months old who drink water containing nitrate above the MCL could become seriously ill and, if untreated, may die, with symptoms including shortness of breath and blue baby syndrome. That standard governs public water systems. A private well is yours to test and yours to maintain.

The practical asymmetry in every one of these cases is causation. Meta says a groundwater study it commissioned found water flows away from the plaintiffs' properties. Amazon denies wrongdoing in the Oregon settlement. Operators arrive with hydrogeology and homeowners arrive with a photograph of brown water, and the fight is about which came first. The only thing that resolves that cleanly is a dated water test from before construction started.

When people ask what the cons of living near a data center are, or ask for the pros and cons, they usually get a marketing brochure on one side and a doomsday post on the other. Here is the sober version.

The cons that are documented: continuous mechanical noise from chillers, cooling towers, air handlers and condenser fans, which does not stop at night; periodic generator testing and, during outages, extended generator running; on some newer AI campuses, on-site gas turbines running as primary power; a construction period of one to several years with heavy truck traffic, dust, and in rocky terrain, blasting; risk to private wells during that construction; large transmission and substation infrastructure and the visual change that comes with it; and cost effects, both to home values where impacts are real and to electricity bills where grid upgrade costs get spread across ratepayers.

The pros that are real: data centers pay substantial property taxes where they are not abated, which can fund schools and roads, and they occupy land that might otherwise become a use with more traffic. They employ relatively few people once operating, so the local job argument is usually weaker than the tax argument.

On home values specifically, the honest answer is that the evidence is contested and highly location dependent, which we lay out in do data centers lower property values. What is clear is the legal mechanics: where noise, water damage or construction harm is real and provable, lost value becomes a damages line inside a private nuisance or property damage case rather than a standalone claim. On electricity, the cost shift is a regulatory fight rather than a lawsuit in most states, and residents can participate in it directly. See data centers and your electric bill and our property value and electric bill claim centers.

The most common one is radiation. The concern usually runs that a building full of servers, switchgear, transformers and transmission lines must be emitting something dangerous. The physics says otherwise, and the distinction is between ionizing and non-ionizing radiation.

WHO addresses electromagnetic fields across the range from 0 to 300 GHz, spanning static electric and magnetic fields, radiofrequency and infrared radiation up through X-rays. The electrical equipment at a data center, and the power lines feeding it, operate at the extremely low frequency and radiofrequency end of that spectrum, which is non-ionizing. The National Cancer Institute states plainly that unlike high-energy ionizing radiation, EMFs in the non-ionizing part of the electromagnetic spectrum cannot damage DNA or cells directly. NCI further reports that no consistent evidence for an association between any source of non-ionizing EMF and cancer has been found in children, and that few studies have reported evidence of increased risk in adults. The International Agency for Research on Cancer has classified extremely low frequency EMFs as possibly carcinogenic to humans based on limited evidence about childhood leukemia at very high exposures, in the range of 0.3 to 0.4 microtesla affecting under one percent of studied populations, while noting that no mechanism has been identified.

That is not a dismissal of resident concerns. It is a redirection. The energy involved in non-ionizing fields is not sufficient to break chemical bonds, which is the mechanism by which ionizing radiation causes damage. If you are worried about a data center next door, the noise at 2 a.m., the generator exhaust plume, the truck traffic and your well are the things worth measuring. The fields are not.

A second category worth naming is the reverse error: industry material that describes these facilities as quiet, invisible neighbors with no local impact. Proposed classes on our tracker already run past a thousand households in Wisconsin and ten thousand residents in Mississippi, and at least one operator has publicly acknowledged a tonal hum reaching a neighboring village. Both the alarmist and the reassuring version fail for the same reason, which is that neither one looks at the measurements.

There is no health-based national setback for a data center. Nothing in federal law establishes a protective distance, and no published study identifies one. Every number that exists comes from a local ordinance or a negotiated agreement, which means it reflects what a county could extract in a deal, not what a researcher concluded.

The most detailed publicly reported example on our tracker is Barber County, Kansas. When the county commission approved Google's Project Helium on August 31, 2026, it did so through a development agreement rather than zoning, because the county has no zoning office. The deal summary requires chillers to sit at least 2,250 feet from non-participating residences and all project infrastructure at least 2,000 feet from the Sharon city limits, with sound capped at 55 dBA daytime and 50 dBA at night at full load, and generators limited to emergencies, maintenance and testing. The full terms and the appeal problem they create are in our Project Helium file.

Set that against the two distances that litigation has produced. The proposed class in the Microsoft Fairwater case covers everyone within a 1.5 mile radius, which is roughly 7,900 feet, more than three times the Barber County chiller setback. The Southaven turbine class exceeds 10,000 people. Plaintiffs' lawyers draw those radii to capture the households they believe were affected, so the numbers are advocacy rather than science, but they are a useful reality check: the people who claim to be harmed are frequently much farther out than the setback line.

Three practical observations follow.

• Distance helps, and the useful distance is larger than most ordinances require. Sound weakens with distance and exhaust disperses, so more separation is genuinely better. But a setback written to protect the nearest house does not describe the edge of the audible zone.
• What is between you and the facility matters as much as how far. Terrain, tree cover, prevailing wind and whether the equipment faces you all change the outcome. Two homes at identical distances can have completely different experiences.
• The number to negotiate for is the sound limit, not just the setback. A distance requirement is static; a nighttime dBA limit measured at the property line is enforceable against how the facility actually operates, and it is the term that gives a resident something to measure.

If you already live next to one. The goal is to convert an experience into a record, because a record is the only thing a regulator, a county board or a lawyer can act on.

• Keep a dated log of the noise. Time, duration, weather, what you were doing, whether it woke you. Consistency over weeks matters more than any single entry.
• Find your local ordinance and its dBA limits and measurement point, then take readings at your property line at the hours that matter. Our noise limits guide shows what these ordinances look like and why A-weighted limits often understate a low frequency hum.
• Pull the facility's air permit. It lists the generator count, emissions tier and permitted run hours, which tells you what is supposed to be happening and lets you flag what is not.
• If you are on a well, test it now even if the water looks fine, and keep the result. If it has already changed, test immediately and photograph everything.
• File complaints in writing with the county, the operator and the state environmental agency. A complaint that exists only as a phone call did not happen.
• See our guide on evidence to document data center harm for the full checklist, and our construction damage center if blasting or dust is the problem.

If you are buying a house, or one is proposed near you. The questions to ask before you sign are boring and cheap, and they are the ones nobody asks.

• Check the county planning and commission agendas and minutes for the last two years for the parcels within a few miles, and look for rezonings, special use permits, development agreements and utility easements. In counties without zoning, look for development agreements specifically, because there may be no zoning file to find.
• Look for large land assemblages, new substation applications and transmission line proposals. Those precede the announcement.
• Ask the utility whether a large load interconnection request exists in the area. Grid capacity is requested long before a project is public.
• If a project is already approved, get the actual conditions in writing: setbacks, dBA limits day and night, the measurement point, generator run restrictions, water source and any decommissioning security. The Barber County terms above are a useful template for what a well drafted agreement contains.
• Visit the nearest comparable operating facility at 2 a.m., not at noon. Every complaint in this category is a nighttime complaint.
• If you are on a well, get a baseline test before closing.
• If you want to oppose a proposal, our guide on how to fight a proposed data center covers the process and the deadlines, which are short.

If a facility near you is already causing harm you can document, a free case review connects you with an independent attorney in our network who handles data center cases in your state, at no cost and with no obligation. What they will want to see is the record described above.