Health concerns: documented, alleged, or unsupported?

DocumentedChronic environmental noise harms health. The World Health Organization’s guidelines, built on decades of evidence (mostly from traffic noise), link sustained exposure above ~53 dB daytime / 45 dB nighttime averages to sleep disturbance, annoyance, and increased cardiovascular risk,³² with 40 dB as the protective nighttime target.³¹ The European Environment Agency attributes roughly 12,000 premature deaths a year in Europe to environmental noise.⁹⁸ This is the mechanism by which a badly sited, noisy facility could harm health — it is not a study of data centers specifically.

DocumentedSome data centers near homes produce noise in the 40–59 dB range and generate real complaints; most produce none. That’s the finding of Virginia’s legislative audit agency across the largest data center market on Earth — and note that 40–59 dB straddles the WHO’s nighttime guidance, which is why siting and design rules matter.¹⁴

Alleged / unresolvedIn Granbury, Texas, more than 40 residents near a Bitcoin mine (with noise readings reported around 85–90 dB) described migraines, vertigo, tinnitus, and ER visits to TIME’s investigators;³⁷ their nuisance lawsuit survived a motion to dismiss and remains pending.³⁶ The noise levels and complaint volume are documented; whether the noise caused the individual diagnoses has not been scientifically or legally resolved. Critically, this was an air-cooled crypto mine 100 yards from homes — the loudest facility class in the industry — not a conventional data center.³⁵

DocumentedDiesel exhaust is a Group 1 carcinogen — the WHO’s cancer agency classified it as carcinogenic to humans in 2012, based on lung cancer evidence from occupational exposure.⁷⁵ Every conventional data center has diesel backup generators.

DocumentedBut the exposure context matters: data center generators are emergency equipment that runs during outages and brief scheduled tests. Federal rules cap non-emergency testing; Virginia permits cap total runtime around 500 hours per year, require modeling to show air standards won’t be exceeded before permits issue, and now set the cleanest engine class (Tier 4) as the default for new data centers.⁷⁶ Virginia operators experienced 0–2 minor outages per site over two years — generators there run almost entirely for maintenance.¹⁴ The legitimate documented concern is cumulative: eastern Loudoun County alone has ~4,700 permitted generators.⁷⁷ A 25–30 MW facility would host on the order of a dozen — a tiny fraction of that — but permit terms (engine tier, hours, testing windows) are worth reading.

Alleged / unresolved The Memphis xAI case — the highest-profile air-quality fight in the country — involves an atypical facility that ran dozens of unpermitted gas turbines as primary power for over a year (documented).⁸² Whether it measurably degraded neighborhood air is genuinely unresolved: city-commissioned testing found no dangerous levels of the pollutants it measured,⁷⁹ critics noted ozone — the pollutant of greatest concern — was never tested,⁸⁰ and independent satellite analyses have reached conflicting conclusions.⁸¹ Nothing public suggests on-site turbine power is contemplated in Pikeville.

Documented One modeling study (UC Riverside/Caltech) estimated air pollution from poweringU.S. data centers — mostly upstream power plants — at ~360 premature deaths in 2023. That’s a grid-wide statistical model of electricity generation, not measured harm from any individual facility.⁷⁸

No supporting evidence Data centers, substations, and power lines produce non-ionizingpower-frequency fields — physically incapable of damaging DNA the way X-rays do. The WHO’s expert review concluded no health consequences from typical low-level exposure have been established; field strength falls off rapidly with distance, and exposure outside a facility fence is comparable to ordinary urban background.⁸³(The oft-cited “possibly carcinogenic” IARC 2B label for magnetic fields is the agency’s weakest category and reflects an unconfirmed statistical association, not established causation.)

DocumentedCooling towers, as an equipment class, are the leading environmental source of Legionnaires’ disease outbreaks when poorly maintained — a real, manageable risk addressed by CDC guidance and ASHRAE Standard 188 water management plans.⁸⁴ No data-center-linked outbreak was found in the records reviewed, and many modern facilities use no cooling towers at all (see the Water page).

Alleged / unresolvedIn Newton County, Georgia, residents 1,000 feet from Meta’s construction site reported wells running dry and sediment in their water; Meta’s commissioned groundwater study found no connection, no pre-construction well survey existed, and a federal inquiry was promised. Unresolved either way.⁸⁵

No supporting evidenceClaims that operating data centers chemically “poison” drinking water: no health-department or court record of chemical drinking-water contamination from an operating data center was found. The documented water issues are about quantity, peak demand, and construction-phase sediment — planning issues, not toxicity.

Documented The recurring, well-documented complaints near data centers are light pollution, multi-year construction traffic, visual impact of industrial-scale buildings, and the cooling hum — all chronicled extensively in Northern Virginia.¹⁴ Operating traffic is light (few employees); construction traffic for 1–3 years is not.

Property values cut against the common assumption. The best independent study — George Mason University, analyzing 2023 home sales — found Northern Virginia homes closer to data centers sold for more, likely because data centers cluster near strong infrastructure and jobs.⁸⁹ No rigorous study isolates the effect of being immediately adjacent to a loud facility, where anecdotal discounts are claimed — genuinely unstudied. Industry-funded analyses (e.g., Mangum Economics, commissioned by data center trade groups) claim large property-tax savings for homeowners; treat those as advocacy with real but unaudited numbers.⁹⁰