Research & studies
Every study and agency report below links to the full text at its original source — so you can read and download the real thing, not our summary of it. We label how directly each one bears on data centers, in plain English.
How to read this library. Direct, data-center-specific health studies are still rare. Much of the strongest science measures the same exposures — constant noise, diesel exhaust, heavy water use — in other settings and applies by analogy. We tag each entry Direct, Modeled, or Analogous so you always know which is which.
Research that looks at data centers specifically. Note the honest state of the field: a 2026 peer-reviewed review concludes there is not yet direct epidemiology measuring health outcomes in people living near a facility. The strongest numbers so far come from emissions and lifecycle models, not from studies of actual neighbors.
Frontiers in Climate · 2026
A peer-reviewed review of what is — and is not — known about data centers and health. It documents the 40–59 dB hum, thousands of diesel backup generators, and large water draws, while stating plainly that direct epidemiological evidence linking facilities to measured health outcomes does not yet exist.
arXiv preprint (Caltech / UC Riverside authors) · 2024
A lifecycle model estimating that, on current trajectories, air pollution tied to U.S. data center electricity and backup generators could carry a public-health cost exceeding $20 billion per year by 2028. These are modeled projections, not measured outcomes near specific facilities.
Read / download the full study (PDF)UC Riverside News (with the think tank Next 10) · 2025
Reporting on UC Riverside research (Shaolei Ren) finding that the public-health burden from California data-center pollution roughly tripled from 2019 to 2023 and could rise another ~72% by 2028 without policy change — driven by diesel-generator nitrogen oxides, fine particulates, and gas-plant emissions. The researchers stress these are modeled estimates made 'in the dark' because facilities disclose little environmental data.
Read the full study at the sourceEnvironmental Health Project (nonprofit) · 2025
A plain-language overview from the Environmental Health Project, a nonprofit, that pulls the concerns together in one place — generator and power-plant air pollution (citing the 2025 modeling projecting ~1,300 premature deaths and over $20B in health costs a year by 2030), heavy water use, light pollution, and noise. It is an advocacy-oriented summary that leans on peer-reviewed and government sources rather than a study in its own right.
Read the full study at the sourceLawrence Berkeley National Laboratory (U.S. Dept. of Energy) · 2024
The Department of Energy's benchmark report on the scale of the build-out. It documents that U.S. data center electricity use reached 176 TWh in 2023 — about 4.4% of the national total — and could hit 325–580 TWh by 2028, and it models the water footprint that scale implies through on-site cooling and power generation. The authoritative reference for how fast the load, and its resource demands, are growing.
Read the full study at the sourceData centers run a constant 24/7 hum. There is no long-term study of that specific hum yet, but there is a deep, high-quality body of science on environmental and traffic noise at comparable levels — the closest analog for what neighbors are exposed to.
Int. Journal of Environmental Research & Public Health (van Kempen et al.) · 2018
The systematic review underpinning the WHO noise guidelines. It found the risk of ischemic heart disease rises about 8% for every 10-decibel increase in road-traffic noise (rated high-quality evidence), with risk climbing continuously from roughly 50 dB and no clear safe threshold.
Journal of the American College of Cardiology (Hahad et al.) · 2019
Explains how chronic noise harms the body even during sleep: it triggers a stress response that raises cortisol and adrenaline, elevates blood pressure, and promotes vascular inflammation — the biological pathway linking persistent noise to heart disease.
Read the full study at the sourceJournal of Exposure Science & Environmental Epidemiology (Münzel et al.) · 2024
A recent synthesis estimating roughly a 3% increase in cardiovascular risk per 10-decibel rise in transportation noise, with significant associations for stroke, heart failure, and cardiovascular death, and measurable effects beginning around 45 dB.
Read the full study at the sourceInt. Journal of Environmental Research & Public Health (Basner & McGuire) · 2018
The companion to the WHO cardiovascular review, focused on sleep. It found sufficient evidence that nighttime transportation noise fragments sleep and raises awakenings and self-reported disturbance — the pathway by which a constant nighttime hum can wear on health over time. The sleep evidence is from traffic and aircraft noise, applied by analogy.
Read the full study at the sourceLarge data centers keep banks of diesel generators on standby and run them for testing. Diesel exhaust is a recognized health hazard; these sources document both the emissions and the underlying harm.
NESCAUM (Northeast States for Coordinated Air Use Management) · 2014
An air-agency analysis of stationary diesel backup generators — the same class of equipment data centers rely on — detailing their NOx and fine-particulate emissions and the disproportionate pollution they create relative to their run-hours.
Read / download the full study (PDF)arXiv preprint · 2025
A modeling estimate that routine generator testing alone at a single large Texas facility can emit on the order of 12 metric tons of NOx per year, worsening ozone in the Dallas–Fort Worth and Houston areas, which are already in federal ozone non-attainment. A single-author preprint, not peer-reviewed.
Read / download the full study (PDF)VCU News (Virginia Commonwealth University) · 2025
Reporting on VCU research finding that backup-generator emissions of CO, NOx, and fine particulates across Northern Virginia's data center corridor rose sharply from 2015 to 2023 — a real-world look at emissions from an actual data center cluster.
Read the full study at the sourceInside Climate News (Arcelia Martin) · 2025
Investigative reporting on the diesel backup generators that keep data centers at 99.999% uptime. It notes the American Cancer Society links diesel emissions to heart and lung disease and cancer, that natural-gas generators emit far less NOx, and that operators largely self-report emissions to Texas and federal regulators — in metro areas that already violate air-quality standards.
Read the full study at the sourceVCU Institute for Sustainable Energy & Environment (Pitt, Suen & Plisko) · 2026
The underlying research behind the VCU coverage. The authors mapped backup-generator air pollution from 138 Northern Virginia data centers and found that, while any single facility is minor, the combined emissions from 100+ facilities are significant — and in some neighborhoods near clusters already rival the exposure from a regional power station. It also examined how that exposure tracks with race, income, and education.
Read the full study at the sourceAmerican Lung Association · 2026
The American Lung Association's annual air-quality report card. The 2026 edition places Dallas, Houston, Brownsville, and McAllen among the nation's most polluted cities — which means the diesel-generator and power-plant emissions tied to new data centers land in Texas airsheds that already carry an unhealthy burden.
Read the full study at the sourcePiedmont Environmental Council / EmPower Analytics Group (Dr. Michael Cork) · 2026
An EPA COBRA modeling study of the permitted on-site power at a Vantage data center in Loudoun County, Virginia — eight gas turbines and 51 diesel generators (~135 MW). It estimates the fine-particle (PM2.5) emissions would cause roughly $53–99 million a year in health damages and 3.4–6.5 premature deaths a year, rising to about $1.6–3.0 billion and 100+ deaths over 30 years. A direct preview of the 'behind-the-meter' gas-and-diesel power model now coming to Texas.
Read the full study at the sourceSouthern Environmental Law Center / EmPower Analytics Group (Dr. Michael Cork) · 2026
A PM2.5 modeling study of xAI's Colossus II / MZX Tech power plant near Memphis: its permitted gas turbines could emit roughly 20 tons of fine particles a year plus precursor pollutants, raising downwind communities' PM2.5 by up to 0.5 µg/m³ — a mortality-risk increase the authors compare to alcohol-impaired driving. The study anchors the NAACP and SELC legal challenge, and is a cautionary precedent for behind-the-meter data-center power in Texas.
Read the full study at the sourceCooling consumes enormous volumes of water and can stress local aquifers and wells. These sources document the consumption and the contamination pathways. Most treat groundwater and discharge risks as potential harms drawn from the broader literature rather than harm measured at a specific facility.
University of Georgia, College of Ag & Environmental Sciences (Field Report TP121) · 2025
A land-grant university field report: medium facilities can use up to 300,000 gallons a day and large ones up to 5 million, much of it drinking-quality; groundwater pumping can lower water tables and affect private wells; and cooling-tower discharge can carry biocides, heavy metals, glycols, and PFAS that are hard to remove.
Read the full study at the sourceAGU Advances (American Geophysical Union) · 2025
A peer-reviewed assessment confirming data centers evaporate on the order of 1–9 liters per kilowatt-hour and that, while the national water footprint is modest, the impact is significant and concentrated in already water-stressed regions — the situation across much of Texas.
Read the full study at the sourcearXiv preprint (UC Riverside / UT Arlington); later in Communications of the ACM · 2023
The widely-cited estimate of AI's water footprint: training a model like GPT-3 in U.S. data centers can directly evaporate roughly 700,000 liters of clean freshwater, and global AI water withdrawal could reach 4.2–6.6 billion cubic meters a year by 2027. Modeled figures, since published in a peer-reviewed ACM journal.
Read / download the full study (PDF)ATSDR, U.S. Department of Health & Human Services · 2021
The federal health agency's profile of PFAS — the 'forever chemicals' that can turn up in cooling-tower discharge. It links PFAS exposure to effects on the liver, immune system, and cholesterol, to raised blood pressure in pregnancy, and to increased risk of some cancers, and explains why they are so hard to remove from water. It establishes why the contamination pathway matters; it does not measure PFAS near a specific data center.
Read / download the full study (PDF)Houston Advanced Research Center (HARC) — Dr. Margaret Cook · 2025
A Texas-focused report: it estimates Texas data centers used roughly 25 billion gallons of water in 2025 (direct and indirect) and projects that could climb to about 399 billion gallons a year by 2030 — on the order of 6.6% of the state's water supply — concentrated around DFW, Houston, San Antonio, Austin, and West Texas. Modeled projections, paired with policy fixes like mandatory water-use reporting for large industrial users.
Read the full study at the sourceTwo of the most-asked-about concerns — the sub-audible 'hum' and effects on animals — have the thinnest direct evidence. We include the best available analogs and label them honestly so you can weigh them fairly.
International Journal of Environmental Research & Public Health · 2021
The closest analog for the low-frequency hum debate. It finds audible turbine sound reliably causes annoyance and sleep disturbance, but concludes that sub-audible infrasound is highly unlikely to cause adverse health effects and that cardiovascular links are not established. A useful counter to overstated 'infrasound sickness' claims.
Annals of Animal Science (Angrecka et al.) · 2023
No study examines cattle raised near a data center, but this peer-reviewed review of the cattle-noise science is directly relevant to Texas ranch and dairy land. It documents that once sound exceeds roughly 80–90 dB, it can disrupt adrenal and hormone function, lower milk yield, raise somatic cell counts, and impair fertility — with discomfort at 90–100 dB. Those documented effects occur at higher levels than a data-center hum, so the link is by analogy, not direct proof.
Guides on this site that help you use the evidence above — to protest a permit, protect your rights, and preserve what matters if a facility is already affecting your land.
Rule of Capture, subsidence, nuisance, quiet enjoyment, and permit protests — in plain English.
Step-by-step on opposing an air permit before it's finalized.
Where to push back on groundwater permits through your local GCD.
Documenting noise, water, and value loss so it counts later.
Bring us what you're seeing — noise logs, a dry well, a stalled home sale. We'll help you understand your options, free and confidential.
Studies are summarized in our own words and linked to their original sources; we do not host third-party publications. Summaries are for general information and are not medical or legal advice.