Are Data Centers Making People Sick? What the Evidence Says About Noise, Air Pollution, EMFs and Water

Article type: Evidence review and public-health assessment

Last updated: October 27, 2026

Abstract

Background: The rapid expansion of artificial-intelligence infrastructure has raised concerns about the effects of data centers on nearby communities. Reported problems include persistent low-frequency noise, generator exhaust, rising electricity demand, water consumption and concerns about electromagnetic fields (EMFs) and electrical power quality.

Objective: To distinguish established environmental risks from emerging hypotheses, examine the strength of the evidence linking data centers to human-health outcomes, and identify practical measures that can reduce potential harm.

Review approach: This editorial evidence review examines relevant scientific reviews, public-health guidance, regulatory information, grid-quality reporting and emerging research available through October 27, 2026. It is not a new systematic review or meta-analysis.

Findings: Environmental noise and combustion-related air pollution are established health concerns when exposure is sufficient. Data centers can create these exposures, but their magnitude depends on facility design, operating conditions and location. Electricity demand and water consumption also create environmental pressures that require appropriate assessment. By contrast, the proposition that electromagnetic fields or so-called dirty electricity explain a characteristic cluster of illnesses among nearby residents has not been established by direct clinical or epidemiological evidence.

Conclusion: Communities should not have to choose between dismissing residents' symptoms and accepting unproven explanations. The appropriate response is independent exposure monitoring, transparent environmental assessments, enforceable operating limits and well-designed studies that measure both exposures and health outcomes.

Keywords: data centers; artificial intelligence; public health; environmental noise; diesel exhaust; air pollution; electromagnetic fields; power quality; water use; environmental justice.

1. Introduction: Why Are Communities Concerned About Data Centers?

Data centers are essential infrastructure for cloud computing, online services and artificial intelligence. Their expansion, however, creates significant demands for electricity, cooling, land and supporting infrastructure. Depending on their location and design, facilities can also introduce noise, emissions from backup generators, additional demand on regional electricity systems and pressure on local water resources.

Residents in some communities have reported headaches, sleep disruption, fatigue, difficulty concentrating, dizziness, tinnitus and a persistent sense of vibration or pressure near data-center facilities. These reports deserve to be investigated rather than dismissed.

However, three separate questions must be answered before drawing a causal conclusion:

Are residents experiencing symptoms or a measurable deterioration in their living environment?
Can a specific exposure from a nearby facility be measured at the places where people live, work or sleep?
Does the available evidence demonstrate that this exposure causes the reported health outcomes?

These questions are related, but they are not interchangeable. A report of symptoms establishes a reason to investigate. A measurement can establish the presence of an exposure. Demonstrating a health effect requires evidence linking the two while accounting for alternative explanations and other relevant factors.

A systematic review published in 2026 identified an important limitation in this field: although environmental pressures associated with data centers have been quantified in several domains, the included evidence did not establish health effects through observed epidemiological or clinical outcomes in exposed populations [1]. The absence of direct evidence does not prove that no effect exists. It means that stronger research is needed.

2. What Does the Evidence Actually Show?

The evidence differs substantially by exposure. The following cards distinguish established environmental risks from data-center-specific health effects that remain uncertain.

1. Environmental Noise

What is established: Sufficiently intense or persistent environmental noise can disturb sleep and contribute to annoyance and stress.

What remains uncertain: The actual exposure levels and long-term health outcomes attributable specifically to individual data centers.

Evidence assessment: Established general health risk; limited direct evidence on data-center-specific health outcomes.

2. Diesel Exhaust and Onsite Combustion

What is established: Diesel particulate matter and nitrogen oxides can adversely affect health.

What remains uncertain: Actual neighborhood exposure from a particular facility, taking account of emissions, operating hours, exhaust dispersion and local conditions.

Evidence assessment: Established pollutant hazards; site-specific risk requires measurement and assessment.

3. Electricity Demand and Regional Air Pollution

What is established: Additional electricity demand can increase emissions, depending on which generators supply the additional load.

What remains uncertain: The precise health burden attributable to data-center electricity demand in each region.

Evidence assessment: Plausible and increasingly quantified; some health estimates are modeled rather than directly observed.

4. Water Consumption

What is established: Some data-center cooling systems consume substantial quantities of water.

What remains uncertain: The consequences for an individual community, which depend on water availability, withdrawals, discharge and competing demand.

Evidence assessment: Environmental pressure is established; local consequences vary.

5. Electrical Harmonics and “Dirty Electricity”

What is established: Electrical harmonic distortion is a recognized power-quality issue that may affect electrical equipment.

What remains uncertain: Whether residential exposure to these disturbances causes the symptoms or diseases attributed to them.

Evidence assessment: Power-quality concerns are distinct from unproven claims of human-health causation.

6. Electromagnetic Fields (EMFs)

What is established: Data centers contain electrical equipment that produces electromagnetic fields.

What remains uncertain: Whether ordinary residential exposures associated with data centers cause the reported symptom cluster.

Evidence assessment: A data-center-specific causal relationship has not been established.

7. Low-Frequency Noise and Infrasound

What is established: Low-frequency sound can contribute to annoyance and sleep-related complaints in some circumstances.

What remains uncertain: The frequency, intensity and duration of exposure required to produce specific health effects near data centers.

Evidence assessment: A legitimate acoustic concern; specific causal claims require stronger evidence.

This distinction is important. Evidence that an exposure can cause harm in general does not automatically establish that a particular facility has exposed a particular household to a harmful dose.

3. Noise, Low-Frequency Sound and Sleep Disruption

Cooling fans, chillers, pumps, transformers and other mechanical equipment can generate persistent sound. Backup generators may add noise during testing and outages, while facilities that use onsite combustion for routine power generation may produce different and more continuous acoustic exposures.

In some circumstances, residents may hear a low hum even when the sound is not especially loud by conventional measures. Low-frequency noise can also be transmitted through building structures, depending on equipment, distance, construction and site conditions.

3.1. What is known about noise and health?

The World Health Organization's environmental noise guidance identifies sleep disturbance, annoyance and several other health outcomes as relevant concerns in environmental noise assessment [2]. A systematic review of observational studies examining low-frequency noise and infrasound in the general population found associations with annoyance, sleep-related problems, concentration difficulties and headaches in some settings, but emphasized that the evidence was limited and methodologically weak, particularly for chronic health conditions [3].

These findings support monitoring low-frequency noise where residents report persistent disturbances. They do not demonstrate that all data centers emit harmful infrasound, that infrasound explains all reported symptoms or that hearing a low hum necessarily indicates tissue injury.

3.2. Low-frequency noise is not the same as infrasound

Sound below approximately 20 Hz is conventionally classified as infrasound. Low-frequency sound can extend into frequencies that people hear, although it may sound more like a hum or vibration than a clearly identifiable noise.

Claims that a large proportion of a data center's emissions consists of medically harmful infrasound require facility-specific acoustic measurements. A description of a sound as a vibration or pressure sensation is not sufficient to identify its frequency or determine the exposure level.

3.3. What should be measured?

An adequate environmental noise assessment should measure sound at residential receptors, including indoors where appropriate, during daytime and nighttime operating conditions. It should consider tonal noise, low-frequency components, vibration and changes associated with generator testing or other operational activities.

Measurements and complaints should be recorded over time. This helps distinguish a persistent equipment-related problem from temporary construction noise, traffic, other industrial sources or unrelated environmental conditions.

4. Diesel Generators, Gas Turbines and Air Pollution

Air pollution is one of the more concrete environmental-health concerns associated with certain data centers.

Many facilities install diesel generators to provide backup power when the grid fails. These systems are normally intended for emergency use and scheduled testing, although the operating pattern varies by facility. Other facilities may use onsite gas turbines or engines to provide some or all of their electricity.

The distinction matters: a standby generator that operates briefly is a different exposure source from a large combustion installation running for extended periods. Installed generator capacity alone does not establish actual emissions or health risk.

4.1. Which pollutants matter?

Relevant pollutants may include:

Fine particulate matter: Small particles can penetrate deep into the respiratory system and contribute to cardiovascular and respiratory harm.
Nitrogen oxides: These emissions contribute to nitrogen dioxide exposure and the formation of ozone and other air pollutants.
Sulfur dioxide and other combustion pollutants: Their significance depends on fuel, equipment, controls and operating conditions.
Greenhouse gases: These contribute to climate change, although their climate effects should be distinguished from immediate neighborhood exposure to local pollutants.

Air-quality regulators explicitly identify backup engines at data centers as sources requiring assessment or permits under applicable rules. The Bay Area Air District, for example, describes diesel particulate matter and nitrogen oxides as pollutants associated with health concerns near these facilities [4]. Washington State's Department of Ecology also provides guidance on diesel-generator emissions and their potential effects on neighboring communities [5].

4.2. What determines a resident's exposure?

The number of installed generators is only one part of the assessment. Important variables include:

Actual operating hours and load, including emergency operation and testing;
Fuel type, engine technology and emissions-control equipment;
Exhaust-stack height, nearby buildings and local meteorology;
Distance to homes, schools, healthcare facilities and other sensitive receptors;
Existing industrial emissions and background air pollution.

Permitted emissions, modeled emissions and measured ambient concentrations are different quantities. A permit may authorize a maximum operating scenario that does not reflect routine conditions, while an emissions model estimates what may occur under specified assumptions. Independent monitoring can help determine whether real-world exposure aligns with those estimates.

Editorial conclusion: The potential health effects of combustion pollutants are not speculative. The unresolved question is the magnitude and distribution of actual exposure around each facility, including cumulative exposure from neighboring sources.

5. Electricity Demand and Pollution Beyond the Data Center

A data center does not need to emit exhaust directly into a neighborhood for its electricity demand to have environmental consequences. Its power may be supplied by a mix of natural gas, coal, nuclear generation, hydroelectricity, wind and solar power, depending on the region and time.

When additional demand requires more fossil-fuel generation, emissions can occur far from the facility consuming the electricity. Conversely, the effect of additional demand depends on which generators respond, how the power system is operated and what clean-energy resources are available.

The International Energy Agency's 2026 analysis projects that worldwide electricity consumption by data centers could rise from approximately 485 terawatt-hours in 2025 to approximately 950 terawatt-hours in 2030 [6]. These are projections, not a guarantee of future consumption, but they illustrate why power supply and grid planning matter.

Researchers are also beginning to quantify the associated air-quality consequences. A 2026 EarthArXiv preprint modeled the air-pollution effects of US electricity demand attributable to data centers and estimated additional particulate pollution and changes in ozone across multiple regions [7]. As a preprint, this work should be treated as emerging, non-peer-reviewed evidence. Model-based estimates should not be presented as observed cases of illness among people who live near individual facilities.

Three separate impacts therefore need to be assessed:

Local impacts: Noise, generator emissions, land use and other effects immediately around a facility.
Regional impacts: Air pollution, grid congestion, electricity prices and water demand associated with the wider power system.
Global impacts: Greenhouse-gas emissions and resource use across the facility's electricity and equipment supply chains.

A responsible assessment considers all three instead of focusing exclusively on the facility boundary.

6. What Does “Dirty Electricity” Mean?

Electrical power-quality problems can include harmonic distortion, voltage fluctuations, transients and other deviations from ideal electrical waveforms. Nonlinear electrical loads and power-conversion equipment can contribute to some of these disturbances.

The phrase dirty electricity is also used in some health discussions to describe high-frequency electrical transients. However, it is not a single standardized measure of residential exposure, and different studies may use different instruments, frequency ranges and metrics.

6.1. What did the Bloomberg analysis find?

In December 2024, Bloomberg published an analysis of power-quality data from approximately 770,000 household sensors, compared with the locations of nearly 1,500 US data centers [8]. It reported that highly distorted power readings were disproportionately concentrated within certain distances of data-center activity.

This finding raises a legitimate engineering question: why was power-quality distortion more common in the identified locations? Possible explanations need to be tested using electrical measurements, grid data and facility-level information.

However, the analysis did not measure residents' symptoms, diagnose disease or demonstrate that data-center exposure caused human-health effects.

6.2. The 8% threshold is not a human-health threshold

The Bloomberg analysis discussed total harmonic distortion (THD) and a threshold of 8% in the context of electrical power quality. Such engineering thresholds concern the characteristics and performance of electrical systems and equipment. They should not be interpreted as a medically validated threshold at which a person becomes ill.

It is also important to distinguish measurements taken at a data center, measurements in a residence and measurements at a point on the electricity network. These do not necessarily represent the same exposure or establish how a disturbance propagates through a particular system.

6.3. Do electrical harmonics cause headaches, brain fog or tinnitus?

A 2010 review of the published literature on high-frequency electrical voltage transients identified methodological problems that prevented a valid assessment of whether these exposures caused adverse health effects [9]. Subsequent research would need to resolve those limitations through better exposure characterization, suitable comparison groups and reproducible outcomes.

Electrical harmonics should therefore be investigated as a power-quality concern where measurements justify it. Claims that these disturbances are responsible for headaches, asthma, tinnitus, neurological disease or other specific illnesses require direct evidence and should not be inferred from correlations between sensor readings and facility locations.

7. Electromagnetic Fields: What Does the Medical Evidence Say?

Data centers contain electrical distribution systems, transformers, server power supplies and other equipment that produce electromagnetic fields. These sources differ from radiofrequency emissions generated by wireless communication equipment, and their characteristics depend on the specific technology and operating conditions.

The broader question of whether exposure to electromagnetic fields can affect health has been studied for decades. The World Health Organization states that current scientific evidence does not confirm adverse health consequences from low-level electromagnetic-field exposure generally, although research into particular exposure types and long-term outcomes continues [10].

7.1. Electromagnetic hypersensitivity and reported symptoms

Some people describe headaches, fatigue, burning sensations, dizziness, difficulty concentrating or other symptoms that they attribute to electromagnetic fields. The symptoms can be distressing and may significantly affect daily life.

However, systematic reviews of blinded or double-blinded provocation studies have not found robust, reproducible evidence that the presence of electromagnetic fields explains the symptoms reported by people who identify as electromagnetically sensitive [11]. These findings do not mean that the symptoms are imaginary. They mean that the proposed cause has not been reliably established.

Symptoms such as fatigue, headaches, sleep disruption, dizziness and tinnitus can have many explanations, including noise exposure, air-quality problems, sleep disorders, medication effects and other medical conditions. The appropriate response is to investigate possible causes without prematurely attributing every complaint to one exposure.

7.2. What would establish a health risk near a data center?

A credible investigation should measure the relevant electromagnetic-field frequencies and intensities at actual residential locations, compare those measurements with applicable exposure guidelines and use appropriate health-study methods. If the hypothesis concerns electrical transients or harmonic distortion rather than conventional electromagnetic-field exposure, researchers must define and measure that exposure separately.

The available evidence does not justify concluding that data centers expose nearby households to medically harmful EMFs or that EMFs explain the reported cluster of symptoms. Neither should this uncertainty be used to dismiss measurable noise, pollution or power-quality problems.

8. Water Use, Cooling and Environmental Justice

Water use is another important concern, particularly in regions where water supplies are limited or competition between households, agriculture, industry and ecosystems is already significant.

Some data centers use cooling towers that consume water through evaporation. Others use different cooling arrangements, and electricity generation may add indirect water demands. A facility's water impact therefore cannot be estimated accurately from its computing capacity alone.

The US Department of Energy identifies water-use effectiveness as one way of assessing data-center cooling performance. Water use should be considered alongside the facility's energy consumption, cooling system and operating conditions [12].

Local environmental review should disclose, where applicable:

Annual water withdrawals and consumption, including seasonal peaks;
The source of water and the degree of local water stress;
Cooling technology and opportunities for reducing potable-water use;
Wastewater discharge and any relevant thermal or chemical impacts;
Contingency plans for drought, restrictions and competing community needs.

Environmental justice is also relevant. Communities hosting large facilities should have access to meaningful consultation, understandable environmental assessments and enforceable protections. The cumulative effects of several facilities in the same area may be more important than the effect of any single facility assessed in isolation.

9. What Can Be Done to Reduce Potential Harm?

Debate over whether data centers are making residents sick should not delay practical measures that reduce established or plausible environmental risks.

9.1. Require independent noise assessments

Developers should measure sound at nearby homes and other sensitive locations under representative operating conditions. Assessments should account for nighttime noise, tonal components, low-frequency sound, vibration and changes during generator operation. Where complaints continue, follow-up measurements should be independently reviewed.

Appropriate setbacks, equipment enclosures, acoustic barriers, quieter fans and careful equipment placement may reduce exposure. Their effectiveness should be verified rather than assumed.

9.2. Make generator emissions transparent

Permits should be accompanied by clear information about generator type, expected operating hours, permitted emissions and pollution-control measures. Actual operating records should distinguish routine testing, emergency operation and any use for routine or grid-support generation.

Where warranted, regulators should assess local air quality and cumulative exposure from nearby industrial sources. Alternative backup technologies, including appropriately designed battery systems or other lower-emission options, should be evaluated on reliability, lifecycle impacts, cost and site conditions.

9.3. Address the electricity system, not just the building

Planning should account for the regional electricity mix, transmission requirements and the additional generation needed to serve the facility. Renewable power, storage, demand flexibility, transmission upgrades and nuclear generation may all contribute depending on the circumstances.

No single energy technology should be described as a guaranteed, universally cheapest or immediately deployable solution without transparent cost, reliability and lifecycle analysis. New reactor designs, including advanced and thorium-based concepts, should be assessed against demonstrated technical performance, licensing requirements, fuel supply, construction schedules and actual project economics.

Onsite power generation may reduce some pressures on the grid, but it is not automatically pollution-free. Its implications depend on the technology used, where emissions occur, the fuel supply, noise, waste heat and whether backup equipment operates only during emergencies or for extended periods.

9.4. Publish water and environmental performance data

Facility operators should disclose meaningful, comparable information on water consumption, cooling performance, emissions and relevant environmental monitoring. Reporting should use consistent definitions, clearly identify estimated versus measured values and explain significant seasonal variation.

9.5. Investigate complaints with measurable evidence

Residents reporting symptoms should have access to a clear complaints process that records the time, location, nature and duration of the problem. Independent investigators should compare complaint records with facility operating logs, acoustic measurements, air-quality data and other plausible exposures.

Medical assessment should proceed on its own merits. Persistent headaches, tinnitus, breathing symptoms, sleep problems or neurological complaints deserve appropriate clinical evaluation, whether or not an environmental cause has yet been identified.

10. What Research Is Needed?

The most important research gap is the absence of robust studies linking well-characterized data-center exposures to observed health outcomes in exposed populations.

Future investigations should move beyond counting complaints or comparing distances on a map. They should establish what people are actually exposed to, how exposure changes over time and whether health outcomes differ in appropriately selected comparison groups.

Priority research areas include:

Prospective community studies: Collect baseline data before a new facility opens, then assess changes in exposures and health outcomes over time.
Personal and residential exposure measurements: Combine environmental monitoring with location, timing and duration of exposure rather than relying solely on distance to a facility.
Acoustic research: Measure low-frequency noise and vibration alongside conventional sound levels, using validated exposure and outcome measures.
Air-pollution studies: Link generator operating records, emissions inventories, meteorology and ambient monitoring to assess real-world exposures.
Power-quality investigations: Establish whether measured electrical disturbances can be attributed to particular sources and whether any proposed health associations are reproducible under rigorous study designs.
Cumulative-impact assessment: Examine multiple data centers, existing industrial sources, traffic and regional energy-generation impacts together.

Research should be conducted transparently, with appropriate ethics oversight, independent analysis, public disclosure of methods and conflicts of interest, and clear separation of prespecified outcomes from exploratory findings.

11. Limitations of This Review

This article is an editorial evidence review rather than a formal systematic review. It does not present a new exhaustive search, pooled effect estimates or a formal risk-of-bias assessment for every study on data centers and health.

Research in this field is changing rapidly. Some emerging estimates rely on environmental models, engineering analyses or preprints that have not completed peer review. Modeled health effects can be useful for identifying potential hazards and guiding policy, but they are not equivalent to observed clinical outcomes among residents.

The absence of direct epidemiological evidence should be interpreted as uncertainty, not proof of safety. Equally, reports of symptoms or correlations between environmental measurements and facility proximity should not be presented as proof of a specific mechanism or disease without additional evidence.

12. Conclusion

Are data centers making people sick? The scientifically responsible answer is that some data-center-related exposures can plausibly affect health, but the overall magnitude of health effects among nearby communities remains insufficiently established.

Persistent noise can interfere with sleep and well-being. Diesel exhaust and other combustion emissions can create genuine respiratory and cardiovascular hazards when exposure is sufficient. Growing electricity demand can increase environmental burdens beyond a facility's immediate boundaries, while water consumption and land-use changes can place pressure on host communities.

These concerns deserve action without requiring an unsupported theory about electromagnetic fields or dirty electricity. Current evidence does not establish that these exposures cause the particular cluster of symptoms reported near data centers, and the power-quality findings reported in household sensors do not demonstrate such a causal relationship.

The way forward is neither blanket reassurance nor alarmism. It is transparent permitting, independent monitoring, responsible facility design, cleaner and more reliable power, water stewardship and rigorous public-health research.

Communities should not have to prove a speculative mechanism before being protected from measurable pollution or excessive noise. Equally, public-health conclusions should be based on reproducible evidence rather than assumptions about what must be causing an individual's symptoms.

References

Pinto A, D'Alterio G, Borlini S, Signorelli C, Pennisi F. Environmental Pressures and Modelled Health Pathways of Data Centers in the AI Era: A Systematic Review and Evidence Map. Environments. 2026;13(10):568. doi:10.3390/environments13100568.
World Health Organization Regional Office for Europe. Environmental Noise Guidelines for the European Region. 2018. WHO guideline.
Baliatsas C, et al. Health effects from low-frequency noise and infrasound in the general population: Is it time to listen? A systematic review of observational studies. Science of the Total Environment. 2016;557–558:163–169. doi:10.1016/j.scitotenv.2016.03.065.
Bay Area Air Quality Management District. Data Centers and the Air District. Updated October 2026. Regulatory guidance.
Washington State Department of Ecology. Data Centers: Air Pollution. Environmental and air-quality guidance.
International Energy Agency. Key Questions on Energy and AI. 2026 update. IEA analysis.
Chen Y, Wang S, Senga JRL, et al. Air quality and health impacts of Data Center electricity demand in the United States. EarthArXiv preprint, 2026. Not peer reviewed at the time of this review. Preprint record.
Nicoletti L, Malik N, Tartar A. AI Needs So Much Power, It's Making Yours Worse. Bloomberg. December 27, 2024. Sensor analysis and methodology.
de Vocht F. "Dirty electricity": what, where, and should we care? Journal of Exposure Science & Environmental Epidemiology. 2010;20:399–405. doi:10.1038/jes.2010.8.
World Health Organization. Radiation: Electromagnetic Fields. WHO evidence overview.
Rubin GJ, Nieto-Hernandez R, Wessely S. Idiopathic environmental intolerance attributed to electromagnetic fields: an updated systematic review of provocation studies. Bioelectromagnetics. 2010;31(1):1–11. PubMed: PMID 19681059.
US Department of Energy. Cooling Water Efficiency Opportunities for Federal Data Centers. Water-use effectiveness and cooling guidance.
A Midwestern Doctor. What Is Making People Near Data Centers Sick (and What Could Fix It)

Editorial disclosure: This review is intended for public education and policy discussion. It does not diagnose environmental illness or establish that a particular facility has caused an individual's symptoms. Site-specific conclusions require appropriate environmental monitoring and, where health effects are alleged, suitable epidemiological or clinical evidence.