The Frequency Nobody Is Measuring: Data Centers, 20 Hz, Heat, and the Hidden Health Cost of the AI Boom
There is a question being asked in communities surrounding large data centers that deserves considerably more scientific attention:
What happens when an industrial facility operates continuously—day and night, month after month—while producing enormous amounts of heat, low-frequency noise, vibration, and electromagnetic energy?
The obvious thing to measure is what we can easily see.
The buildings.
The electricity consumption.
The cooling towers.
The water usage.
The temperature of the air coming out of the facility.
And, when residents complain about noise, the decibel level.
But there is another part of the story that is much harder to see.
It is measured in hertz rather than miles per hour, frequency rather than volume, and acoustic pressure rather than ordinary decibels.
And that raises a troubling question:
Are communities measuring the wrong characteristics of data-center noise?
A data center can be operating within a seemingly acceptable A-weighted noise limit while still producing substantial low-frequency energy. Research into commercial HVAC and server environments has specifically identified low-frequency noise below 250 Hz and found disparities between A-weighted and C-weighted measurements. The researchers concluded that conventional A-weighted regulations can fail to adequately capture low-frequency noise.
That distinction matters because 20 Hz is not ordinary background noise.
It sits directly at the conventional boundary between low-frequency sound and infrasound. Human hearing becomes dramatically less sensitive as frequency falls, but that does not mean the sound ceases to exist—or that it necessarily becomes irrelevant simply because an A-weighted meter reports a relatively modest number. Research reviewing human perception at low and infrasonic frequencies confirms that people can perceive very-low-frequency sound when the sound-pressure level is sufficiently high.
And this is where the data-center question becomes much more interesting.
The 20-Hertz Question
Imagine standing several miles from a massive computing facility.
You cannot see the cooling equipment.
You cannot hear a conventional high-frequency mechanical roar.
There may not even be an obviously loud sound.
Yet the facility is operating continuously.
Thousands upon thousands of processors are generating heat. Cooling systems are removing that heat. Fans, pumps, chillers, compressors and other mechanical equipment are cycling or operating continuously. Electrical infrastructure is carrying enormous loads.
Some of the machinery inevitably produces acoustic energy.
Some of that energy occurs at low frequencies.
And the lower the frequency, the longer the wavelength.
At 20 Hz, the wavelength in air is roughly 17 meters (56 feet) under ordinary atmospheric conditions.
That is an enormous physical wavelength compared with the dimensions of many ordinary buildings.
The question is therefore not simply:
“Can I hear it?”
The better questions are:
What is the sound-pressure level at 20 Hz?
How does it change with distance?
How does terrain affect it?
How do temperature gradients and atmospheric conditions affect propagation?
Does the facility generate persistent tonal components?
Are multiple machines producing the same or harmonically related frequencies?
And what happens to residents who are exposed continuously rather than intermittently?
Those are measurable scientific questions.
They are also questions that standard A-weighted community-noise measurements may not fully answer.
Why 20 Hz Deserves Special Attention
The distinction between frequency and loudness is critical.
A sound at 20 Hz and a sound at 2,000 Hz can have very different physical and perceptual characteristics even if a conventional meter reports similar overall sound levels.
Low-frequency sound also interacts differently with buildings and the environment. Researchers studying low-frequency noise have repeatedly identified the importance of considering frequency weighting and frequency spectra rather than relying exclusively on a single A-weighted number.
A 2023 review of experimental research on infrasound and low-frequency noise specifically examined cardiovascular effects and sleep disorders, along with annoyance and other health outcomes.
A companion review of epidemiological research likewise examined cardiovascular outcomes and sleep-related effects. The authors emphasized that the available evidence remains limited and methodologically challenging—but that low-frequency noise deserves substantially more research attention.
That last point is important.
The scientific question is not settled.
But “not settled” does not mean “not worth measuring.”
It means we need better measurements.
The Heart May Be Part of the Story
This is where the conversation becomes considerably more serious.
Environmental noise is increasingly being investigated as a cardiovascular risk factor. A 2025 umbrella review examining 20 meta-analyses reported associations between environmental noise exposure and hypertension, atrial fibrillation, coronary heart disease and ischemic heart disease, along with evidence concerning stroke. The authors concluded that the evidence supports noise exposure as a potential cardiovascular risk factor, while also distinguishing between different cardiovascular outcomes.
Another systematic review and meta-analysis examined 133 primary studies involving blood pressure, hypertension, heart rate, cardiac arrhythmia, vascular resistance and cardiac output. It found signals suggesting cardiovascular effects from higher noise exposure, although the authors rated the overall certainty of evidence as very low because of substantial methodological limitations and differences between exposure sources and populations.
That qualification should not be buried.
It should be highlighted.
Because it tells us exactly what the next generation of data-center research should investigate.
Not merely:
“Is the data center under 55 dBA?”
But:
What frequencies are present?
At what sound-pressure levels?
For how many hours per day?
At what distances?
Under what atmospheric conditions?
What happens during nighttime temperature inversions?
What happens to sleep?
What happens to heart rate and blood pressure?
And what happens when exposure continues for years?
The Sleep Problem
Sleep may be one of the most important pathways through which chronic low-frequency environmental noise could affect health.
A person does not have to consciously wake up and say:
“That was the data center.”
A persistent acoustic stimulus can potentially disturb sleep architecture, increase annoyance, alter perceived restfulness, or contribute to repeated micro-arousals without producing a dramatic conscious awakening.
Research on low-frequency noise has identified associations with sleep-related problems, headache, concentration difficulties and annoyance, although systematic reviews emphasize that the epidemiological evidence remains limited.
That creates an important public-health question for communities living near continuously operating industrial facilities.
A factory that operates eight hours a day is one exposure.
A facility operating 24 hours a day, seven days a week is another.
And an environment in which multiple large facilities operate simultaneously could represent something else again.
The Sound May Be Low. The Exposure Is Not.
One of the most misleading assumptions about industrial noise is that if it is not obviously loud, it cannot be significant.
But low-frequency noise presents a different measurement problem.
Research involving heat pumps and ventilation systems found greater annoyance and disturbances of rest and concentration among people exposed to low-frequency noise compared with controls exposed primarily to mid-frequency noise.
And a 2024 study examining commercial HVAC systems—equipment conceptually relevant to large computing facilities—found that HVAC and server systems can dominate low-frequency noise below 250 Hz and specifically highlighted the disparity between A- and C-weighted measurements.
That is an extraordinary detail when considering modern data centers.
The very equipment required to keep the computers operating—fans, cooling systems, pumps and HVAC infrastructure—is also capable of becoming a source of low-frequency environmental noise.
The cooling system is not merely an accessory to the data center.
It is part of the acoustic footprint.
And Then There Is the Heat
The noise question should not be confused with the thermal question.
A 20-Hz acoustic wave does not, by itself, provide a scientifically established mechanism for raising atmospheric temperature across a seven-mile radius.
That claim would require evidence.
But there is a completely different and very real thermal question surrounding data centers:
How much waste heat is being released, where is it released, how is it transported, and how far can measurable thermal effects extend under particular atmospheric conditions?
That question can—and should—be studied independently of acoustics.
A large computing facility consumes enormous quantities of electrical energy. Ultimately, nearly all of that electrical energy becomes heat somewhere in the system.
The heat does not simply disappear.
It has to be transferred into the surrounding environment through cooling systems, exhaust air, cooling towers, water systems and other heat-rejection mechanisms.
The resulting thermal plume depends upon facility design, weather, wind speed, atmospheric stability, humidity, terrain, height of discharge and other variables.
Under certain atmospheric conditions, heat and pollutants can travel considerable distances.
That does not mean a data center creates a uniform seven-mile-radius temperature increase.
It means the thermal plume is a dispersion problem, and it should be measured as one.
Watch: What Happens When Low-Frequency Sound Meets the Human Body?
The science of low-frequency sound becomes much more compelling when we move beyond the question of whether we can hear it. This video explores the relationship between very-low-frequency sound, infrasound, and the human body—providing additional context for why researchers continue to investigate potential effects on sleep, stress, cardiovascular function, and overall well-being.
As data centers bring increasingly powerful cooling and mechanical systems into communities, understanding what exists below the familiar range of everyday noise measurements becomes an important part of the conversation.
Could a 20-Hz Signal Travel Seven Miles?
Yes, a 20-Hz sound wave can physically propagate several miles.
That is not the same thing as saying that a harmful 20-Hz exposure will remain at the same level seven miles away.
Those are two completely different claims.
Sound intensity generally decreases with distance, while atmospheric absorption, ground interaction, reflections, terrain and meteorological conditions modify the result. Very-low-frequency sound is comparatively persistent because atmospheric absorption is much weaker than it is at higher frequencies.
But the scientifically meaningful question is not:
“Can 20 Hz travel seven miles?”
The answer to that is essentially yes.
The meaningful question is:
“At seven miles, what is the 20-Hz sound-pressure level above the local background, and does it remain measurably attributable to the facility?”
That requires field measurements.
And those measurements should be made under different atmospheric conditions.
Watch: Understanding Low-Frequency Sound and Infrasound
Before continuing, this video provides useful visual and audio context for understanding low-frequency sound, infrasound, and why frequencies near 20 Hz require different considerations than ordinary audible noise.
Temperature Could Change the Acoustic Footprint
Here is where reports of residents noticing changes as temperatures rise become scientifically interesting.
Temperature does not simply determine whether a sound exists.
It changes the speed of sound and contributes to atmospheric gradients that can influence how sound propagates.
More importantly, the atmosphere is rarely uniform.
Temperature can vary with altitude.
Wind can vary with altitude.
Humidity can vary.
Atmospheric stability can change.
Those gradients can bend acoustic energy.
This is why a noise measurement taken on one afternoon cannot necessarily tell us what residents will experience at 2 a.m. on another night.
A sophisticated investigation would therefore record:
-
temperature at multiple heights;
-
humidity;
-
wind speed;
-
wind direction;
-
atmospheric pressure;
-
atmospheric stability;
-
A-weighted sound level;
-
C-weighted sound level;
-
Z-weighted sound level;
-
narrowband frequency spectra;
-
20-Hz sound-pressure levels;
-
octave and one-third-octave bands;
-
tonal components;
-
and measurements at multiple distances from the facility.
Then repeat those measurements during different seasons.
That would begin to answer the question communities are actually asking.
The Measurement Gap
This may ultimately be the most important issue.
A community is told:
“The facility complies with the noise ordinance.”
But what exactly was measured?
If the ordinance measures only A-weighted sound, the answer may be incomplete.
A-weighting was designed around the frequency sensitivity of human hearing and strongly discounts low-frequency energy.
C-weighting is much flatter through the low-frequency range.
Z-weighting is intended to be essentially unweighted.
None of these measurements alone tells the entire story.
A frequency spectrum tells us something different again.
And an infrasound measurement tells us something different again.
The distinction is not theoretical.
Researchers studying low-frequency noise have explicitly documented disparities between A- and C-weighted measurements and argued for better inclusion of low-frequency components in noise regulations.
That means a regulation can be technically satisfied while still leaving unanswered questions about the low-frequency acoustic environment.
The Cardiovascular Question Cannot Simply Be Dismissed
There is another reason this deserves attention.
The cardiovascular system is exquisitely sensitive to changes in the autonomic nervous system, sleep, stress and environmental conditions.
We already know that environmental noise research has found associations with cardiovascular outcomes.
We also know that low-frequency and infrasonic exposure is sufficiently unusual that researchers continue to investigate whether it can produce physiological effects distinct from ordinary audible noise.
A 2023 review specifically examined cardiovascular effects of low-frequency noise and infrasound.
A separate systematic review of observational studies identified associations between low-frequency noise and annoyance, sleep problems, concentration difficulties and headache, while emphasizing the scarcity and limitations of epidemiological evidence regarding chronic disease.
The responsible conclusion is therefore neither:
“20 Hz causes heart disease.”
nor:
“20 Hz cannot affect the cardiovascular system.”
The responsible conclusion is:
There is enough evidence about environmental noise and cardiovascular health—and enough uncertainty specifically surrounding low-frequency exposure—to justify measuring and studying the frequency spectrum rather than dismissing it.
The Data Center Should Be Treated Like an Industrial Acoustic Environment
This is perhaps the biggest conceptual change required.
A modern hyperscale data center is not simply an office building containing computers.
It is an enormous industrial energy-conversion system.
Electricity enters.
Computational work occurs.
Heat is generated.
Cooling infrastructure removes the heat.
Fans, pumps, compressors, chillers and other mechanical systems operate.
Air and water move.
Electrical equipment operates.
And the facility continues doing this around the clock.
That makes its environmental footprint multidimensional.
There is the electrical footprint.
The water footprint.
The carbon footprint.
The thermal footprint.
And the acoustic footprint.
The acoustic footprint deserves to be measured with the same sophistication applied to the other four.
What a Real Community Study Should Measure
If a community believes a data center is producing an unusual low-frequency environment, there is a straightforward way to move the discussion from anecdote to evidence.
Establish monitoring stations at several distances.
For example:
500 feet
1,000 feet
2,500 feet
1 mile
2 miles
5 miles
7 miles
At every location, continuously record the acoustic spectrum.
Not merely dBA.
Record dBC and Z-weighted levels, together with narrowband frequency data.
Then isolate the 20-Hz component.
If a strong 20-Hz peak exists near the facility, determine whether it decreases with distance according to expected propagation behavior.
Then compare it against the background level when the facility is operating normally and, if possible, during maintenance periods when particular systems are shut down.
At the same time, measure atmospheric conditions.
Temperature.
Wind.
Humidity.
Pressure.
Atmospheric stability.
Then correlate the measurements.
If residents report symptoms or sleep disturbance, those observations can be recorded separately rather than automatically attributed to the sound.
That is how an emotionally charged controversy becomes a scientifically testable environmental-health study.
What We Know—and What We Do Not
We know that low-frequency noise exists in industrial and urban environments.
We know that HVAC systems and computer/server infrastructure can contribute to low-frequency noise.
We know that humans can perceive sufficiently intense low-frequency and infrasonic sound.
We know that environmental noise is associated in the broader literature with cardiovascular outcomes, including hypertension and other cardiovascular conditions.
We know that researchers have investigated low-frequency noise and infrasound in relation to cardiovascular effects and sleep.
We know that the epidemiological evidence specifically connecting chronic low-frequency exposure to serious disease remains limited and methodologically difficult.
And we know that the acoustic environment changes with weather and atmospheric conditions.
What we do not yet know is whether a particular data center is producing a biologically significant 20-Hz exposure at seven miles.
That is the question that needs to be measured rather than assumed.
The Question We Should Be Asking
The debate over data centers has become dominated by enormous numbers.
Megawatts.
Gallons.
Acres.
Buildings.
Jobs.
Tax revenue.
Artificial intelligence.
But another number may deserve a place in the conversation:
20 Hz.
Because if a facility is generating persistent low-frequency acoustic energy, and if that energy is propagating beyond the property boundary, then the question is no longer simply whether the facility is “loud.”
The question becomes:
What frequencies are entering the surrounding community?
And once that question is asked, another follows:
What happens to people who live there for years?
Especially people who are sleeping there.
Children.
Older adults.
People with cardiovascular conditions.
People who already suffer from migraines or sleep disorders.
People who report unexplained nighttime disturbances.
People who say they can feel or hear a persistent low-frequency hum even when conventional noise measurements tell them the environment is supposedly acceptable.
Those experiences should not automatically be accepted as proof of harm.
But neither should they automatically be dismissed.
They are signals.
And signals can be measured.
The Next Frontier of Data-Center Environmental Research
The next generation of environmental assessments should move beyond the question:
“How many decibels?”
and begin asking:
“Which frequencies, at what levels, for how long, and under what atmospheric conditions?”
That means measuring the spectrum.
It means looking below 100 Hz.
It means looking specifically at the 20-Hz region.
It means distinguishing audible low-frequency noise from true infrasound.
It means measuring C-weighted and Z-weighted sound levels rather than relying exclusively on dBA.
It means studying nighttime conditions.
It means examining thermal plumes separately from acoustic propagation.
And it means conducting long-term health research rather than assuming that compliance with a conventional noise ordinance automatically means the environment is harmless.
The technology industry has built machines capable of processing an extraordinary portion of the world's information.
It is time for environmental science to build an equally sophisticated picture of what those machines are doing to the places around them.
Because the sound nobody hears may still be measurable.
The heat nobody sees may still be measurable.
And the health effects nobody has yet connected to a source may eventually become measurable too.
The question is not whether we should be afraid of 20 Hz.
The question is whether we are measuring it carefully enough to know what it is doing.
Research Worth Reading
Low-frequency noise and cardiovascular health
Impact of infrasound and low frequency noise on human health and well-being — Part I: Experimental studies — Reviews experimental evidence, including cardiovascular effects and sleep disorders.
Impact of infrasound and low frequency noise on human health and well-being — Part II: Epidemiological studies — Reviews epidemiological evidence concerning cardiovascular effects, sleep and other health outcomes.
The relationship between noise pollution and cardiovascular diseases — umbrella review — Examines evidence across 20 meta-analyses concerning hypertension, atrial fibrillation, coronary heart disease, ischemic heart disease and stroke.
Impact of Noise Exposure on Risk of Developing Stress-Related Health Effects Related to the Cardiovascular System — Systematic review/meta-analysis examining blood pressure, hypertension, heart rate, arrhythmia and other cardiovascular measures.
Low-frequency sound and human perception
Hearing at low and infrasonic frequencies — Reviews human perception of sound from 20–200 Hz and below 20 Hz.
What is infrasound? — Reviews the definitions and physiological/perceptual issues surrounding infrasound.
Low-frequency noise, sleep and general health
Health effects from low-frequency noise and infrasound in the general population — Systematic review of observational studies examining annoyance, sleep-related problems, concentration and headache.
The prevalence of annoyance and effects after long-term exposure to low-frequency noise — Examines long-term residential exposure to low-frequency noise from heat-pump and ventilation installations.
Data-center/HVAC low-frequency noise
Impact of low-frequency noise on a residential zone due to commercial HVAC systems — Particularly relevant because it examines commercial HVAC systems and identifies server/HVAC equipment as contributors to low-frequency noise below 250 Hz, including differences between A- and C-weighted measurements.
Infrasound and emerging research
Infrasound — implications for human medicine — Reviews current medical understanding and emphasizes that perception and physiological effects depend strongly on sound-pressure level and frequency spectrum.
Infrasound in Biology and Medicine — Insights into Mechanisms, Health Outcomes and Research Perspectives — A newer review examining proposed mechanisms, health outcomes and research gaps.
The unanswered question
The strongest conclusion is not that 20 Hz causes a seven-mile temperature increase. There is currently no sound scientific basis for making that claim.
The stronger—and more provocative—question is this:
If a hyperscale data center is producing persistent low-frequency acoustic energy, while simultaneously releasing enormous quantities of waste heat, can we adequately understand its environmental and health footprint without measuring both the thermal plume and the low-frequency acoustic spectrum over distance and time?
Right now, that is a question worth investigating.
