Bright
← Data CentersEVERGREEN GUIDE / Infrastructure · Public services · Evidence

Why do data centers hum—and how can that be reduced?

Take the disturbance seriously, identify the source and path, then verify the promised control after startup.

Primary sources ↓ · Limitations ↓

SHORT ANSWER

Start here.

A data center can create a persistent hum, but community impact is not inevitable and cannot be inferred from the building type alone. Fans, cooling equipment and transformers may operate continuously or vary with load; generators add episodic testing and emergency sound. What reaches a home depends on equipment, orientation, distance, terrain, weather, other facilities, background sound and time of day. Good projects model the full spectrum before construction, choose and place quieter equipment, use enclosures, silencers or correctly designed barriers, control routine testing, and repeat calibrated measurements after startup with a remedy if predictions were wrong.

Bright publication
Source review
Reading question
What source is operating, what spectrum reaches the receptor, and what evidence shows the mitigation worked?

INTERACTIVE EXPLANATION

Follow sound from source to neighbor.

Select a source and a control. Distance, operating state, terrain, weather, background sound and other facilities still shape what reaches a home.

Choose a source
Choose a control
01Fans + cooling02DISTANCE · TERRAIN · WEATHER03Siting + orientation04NEIGHBOR

Fans + cooling

Character: continuous or load-dependent broadband and tonal sound.

Increase separation where feasible and place openings or equipment so the strongest path does not face a sensitive receptor.

Verification: model the spectrum and operating condition before construction, then measure the same receptor after startup. This diagram does not predict a decibel level.

CLAIMS EXAMINED

Test the exact claim.

MisleadingFACILITY-WIDE GENERALIZATION

“All data centers hum loudly around the clock.”

Some mechanical and electrical sources are continuous and some residents report serious disturbance, but Virginia JLARC found a large majority of facilities did not generate complaints because of siting or design. Generator testing and emergencies are also different from continuous cooling or transformer sound.

What would change this assessment: A representative facility sample with comparable full-spectrum measurements, operating states, siting conditions and complaint outcomes.

Sources: Virginia JLARC: Data Centers in Virginia, Part 2 · Loudoun County: Data Centers—Noise and Air Quality Concerns

UnsupportedORDINANCE COMPLIANCE VS. DISTURBANCE

“If a project meets its dBA limit, it cannot disturb neighbors.”

A-weighting compresses a spectrum into one number and downweights low-frequency energy. Tonal character, persistence, nighttime background and cumulative facilities can matter even when sound is below a local limit or far below a hearing-damage threshold. Compliance answers a legal test; it does not establish absence of disturbance.

What would change this assessment: A local standard and study that cover the relevant receptor, full spectrum, tonal character, background, nighttime conditions and source attribution, plus post-start evidence showing no material increase or complaint pattern.

Sources: Virginia JLARC: Data Centers in Virginia, Part 2 · Loudoun County: Data Centers—Noise and Air Quality Concerns

UnsupportedMEASUREMENT CLAIM

“A phone sound-meter reading or recording proves the facility violates a noise rule.”

A phone can document when a sound is noticed, but built-in microphones and apps vary and do not by themselves establish calibration, frequency response, weather, background, equipment state or source contribution. Compliance work needs calibrated instruments and a documented protocol.

What would change this assessment: A calibrated professional measurement at the required receptor with pre- and post-calibration, operating conditions, weather, background and the ordinance’s specified metrics.

Sources: Evaluation of smartphone sound measurement applications · East Brandywine Township data-center noise requirements

UnsupportedMITIGATION CLAIM

“Trees, one wall, or liquid cooling will eliminate data-center noise.”

A barrier works only when height, length, mass and source-to-receptor geometry block the path; low frequencies can bend around it. Vegetation alone is not a dependable acoustic barrier, and liquid cooling does not remove transformers, pumps, outdoor heat rejection or generators.

What would change this assessment: A site-specific spectral model followed by comparable post-start measurements at the same receptors and equipment states.

Sources: FHWA noise barrier design principles · Virginia JLARC: Data Centers in Virginia, Part 2

DOCUMENTED SOLUTIONS

What helps—and under which conditions.

LOCAL ORDINANCE · PENNSYLVANIA · PRE/POST VERIFICATION

East Brandywine’s measure-build-remeasure framework

The township requires qualified-professional preconstruction work and an as-built study within 180 days. The test must document baseline, equipment and weather, include peak cooling and generators under load, address dBC and tonal sound, and trigger corrective action and retesting when required.

Conditions and limits: This is a strong accountability design, not proof that a particular facility has achieved a reduction. Its numeric limits are local policy, not universal health thresholds or a universal setback.

Sources: East Brandywine Township data-center noise requirements

ADOPTED RULE · VIRGINIA · AUGUST 2026

Henry County’s worst-case and annual-test rule

The county requires third-party modeling of peak cooling and routine generator testing, a tonal-noise prohibition, post-construction testing within 90 days, annual testing, and weekday limits for routine generator tests.

Conditions and limits: No data center was operating under the new framework when adopted. The 1,000-foot setback and 50 dBA property-line limit are local rules, not universal safe distances or proof of performance.

Sources: Henry County adopts data-center regulations

OPERATING EQUIPMENT · VENDOR CASE · INDEPENDENT RAW REPORT UNAVAILABLE

A supplier-reported chiller treatment in Phoenix

A mitigation supplier reports that an outside acoustical consultant measured an average 6.8 dB(A) reduction around one treated air-cooled chiller before the treatment was applied to 48 units.

Conditions and limits: The data center is unnamed, the raw consultant report and receptor-level results are not public, and the measure is around a test chiller—not demonstrated change at neighboring homes. Treat it as source-level supplier evidence only.

Sources: Phoenix data-center chiller mitigation case study

Separate continuous sources from episodic operation

Cooling towers, chillers, dry-cooler fans, pumps, air handlers and transformers can create continuous or load-varying sound. Backup generators are normally a different operating case: routine testing, emergency operation, and in some markets a defined grid call. Construction noise is another temporary phase and should not be folded into an operating-level claim.

An equipment reading inside the site is not the sound level at a home. Sound changes with distance, source height and direction, intervening buildings and terrain, ground conditions, wind and temperature gradients. Multiple nearby campuses may also combine at the receptor.

Sources: Virginia JLARC: Data Centers in Virginia, Part 2 · Loudoun County: Data Centers—Noise and Air Quality Concerns

One A-weighted number cannot describe every hum

dBA is useful and widely regulated, but it downweights lower frequencies. A narrowband transformer or fan tone can be more noticeable than broadband sound at a similar overall level. A quiet nighttime background can also make a constant source more salient without moving it near a hearing-damage threshold.

JLARC reported complaint-triggered measurements of 40–59 dBA in selected Virginia cases. Those readings were total sound, not isolated data-center contribution, and were not representative of every facility. They show why the exact sampling context must stay beside the number.

Sources: Virginia JLARC: Data Centers in Virginia, Part 2 · Loudoun County: Data Centers—Noise and Air Quality Concerns

Design the control around the actual path

Better siting and orientation can keep high-output equipment and openings away from sensitive receptors. Quieter equipment, variable-speed operation, acoustic enclosures, intake and exhaust silencers, treated louvers and correctly dimensioned barriers can reduce specific paths. Routine generator testing can be scheduled for daytime where local rules allow.

None of those measures is a universal guarantee. Lower-frequency sound is harder to block; barriers need adequate geometry; operating controls must preserve safety and reliability; and moving one source can expose another receptor. Model normal operation, peak cooling, routine generator testing and a clearly defined emergency case instead of quoting one best-case point.

Sources: FHWA noise barrier design principles · East Brandywine Township data-center noise requirements

Verification is the difference between a promise and a solution

Before construction, record the baseline at relevant property lines and homes, including daytime, nighttime and octave-band conditions. Model manufacturer sound-power spectra with terrain, buildings, barriers, weather assumptions and simultaneous equipment states. After startup, return to the same receptors with calibrated instruments and documented load conditions.

A useful rule also names the remedy: operating modification, equipment treatment, temporary curtailment where lawful, a corrective plan and a retest. Publish enough of the report for residents to compare the prediction with the measured result. A phone recording remains valuable as a time-and-experience log, not as the final compliance instrument.

Sources: East Brandywine Township data-center noise requirements · Evaluation of smartphone sound measurement applications

FOR A LOCAL PROJECT

What a resident should ask for.

  1. What are the baseline daytime and nighttime levels at the nearest homes and property lines, including octave bands and tonal character?
  2. Which sources operate continuously, vary with computing load, or occur only during testing or emergencies?
  3. Does the model include peak cooling, generators under load, simultaneous equipment and cumulative nearby facilities?
  4. What manufacturer sound-power spectra, terrain, weather and barrier assumptions were used?
  5. Which controls are binding: equipment limits, placement, enclosures, silencers, barriers, test hours, or operating conditions?
  6. When will post-start measurements occur at the same receptors, and will the full report be public?
  7. What corrective action, deadline, retest and enforcement process applies if predictions are missed?

What this guide does not establish

  • Noise impact is site-specific; no universal safe setback follows from the reviewed evidence.
  • Complaint-triggered measurements are not representative facility averages or source-isolated readings.
  • Local dBA and dBC limits are policy choices, not universal health or disturbance thresholds.
  • The only located operating before/after mitigation figure is supplier-reported and lacks a public raw consultant report and residential-receptor outcome.
  • Bright did not perform acoustic measurements or reproduce any project model.

Primary sources

Source-reviewed 2026-09-19. Source review means Bright compared this explanation with the linked records; it is not independent replication or expert peer review by Bright.

Continue into the evidence