Data-center noise can be engineered down — but it is not a solved problem where people live

The question asked: what solutions do data centers have for the sound and noise they generate, is this a solved problem, what are the leading-edge fixes, and how are operators actually responding? This report draws on a refined evidence set of 10 normalized noise-control technologies, 39 extracted records of documented community disputes (covering seven distinct facilities), and 54 policy and measurement requirements — live-web research performed 2026-08-29. The short answer: a well-sited greenfield facility can meet a defensible acoustic target with mature methods, but low-frequency tonal hum, dBA-only ordinances, and complaint-driven retrofits keep the community problem open.

The noise-control stack and its maturity frontier

Sound travels equipment → transmission path → property line → governance. Each card is one documented control or dispute; hover for evidence, caveats, and the source. Orange bars mark quantified attenuation claims (vendor or pilot figures, not independently validated).

commercially deployed pilot / emerging maturity not stated in source live community dispute governance requirement

What the visual cannot say alone

Compliance and complaint coexist: Loudoun County zoning checks never recorded Vantage II above the 55 dB average limit while a resident measured 70–80 dB in hot weather — the ordinance measures only broadband dBA, with no tonal or low-frequency rule. loudounnow.com

Taller walls are not the answer at low frequency: Atlas Power's roughly $5 million barrier in Williston, ND was reported functionally useless against a dominant 118 Hz blade-pass hum measured at 60–95 dB at property lines. sonoranthinktank.substack.com

Old ordinances miss the machine: Prince William County's noise rule, written more than 30 years ago, exempts air-conditioning units — so residents' claims that AWS noise tops the 60 dB limit are, in activists' words, largely moot. apnews.com

The pattern is retrofit, not prevention: shrouds at AWS, fan fixes at Microsoft Mount Pleasant, mufflers at CyrusOne Chandler all followed complaints, lawsuits, or ordinance action — and in each case reporting shows reduction, not elimination, or disputed adequacy. businessinsider.com

Not solved, because…

Broadband dBA limits can be met while an intrusive low-frequency tone persists; most ordinances have no octave-band or tonality rule.
Retrofits come after commissioning: shrouds, fan changes, and barriers followed protests and lawsuits, and barriers can fail outright below ~200 Hz.
Operators and residents disagree on where and how to measure — Amazon, Vantage, and Hyperscale Data all dispute resident or municipal readings.
Nighttime ambients, temperature inversions, cumulative campuses, and generator tests make a single daytime compliance test unrepresentative.
Some ordinances exempt the loudest equipment class (HVAC) entirely.

What good rules now demand

Third-octave 10–200 Hz and G-weighted metrics with a per-band ceiling over baseline — not dBA alone. fowlerfinnlab.com
Independent, applicant-funded studies pre- and post-construction under worst-case load — now required in Fairfax County. fairfaxcounty.gov
Tonal penalties of +3 to +6 dB when a hum, whine or buzz is audible. novascotia.ca
Continuous or seasonal boundary monitoring and a complaint trigger that forces re-testing at the operator's expense.
Setbacks sized from low-frequency propagation — on the order of 1,500+ feet for the largest facilities.

Six disputes that define the pattern

Representative cases from 39 extracted records (duplicate reporting of the same facilities collapsed). Complaint → operator or government action → what stayed open.

Every control in the evidence set

Control	Stage	Maturity	Evidence / claimed outcome	Limitation noted	Source

Greenfield checklist

Model acoustics before permitting under full-buildout, worst-case nighttime load; commission independently. fairfaxcounty.gov
Buy quiet at source: liquid cooling where feasible, low-speed large fans, magnetic-bearing chillers, aisle containment. techtarget.com
Set generous setbacks from homes, cluster equipment centrally, orient openings away from residential interfaces. fowlerfinnlab.com
Schedule generator tests weekday daytime, never simultaneously; monitor the boundary continuously with a public complaint trigger.

Retrofit checklist

Diagnose the spectrum first — find the tone (CyrusOne traced its hum to chilling fans before adding mufflers). heatmap.news
Treat the source before the path: fan speed, VFDs, balancing, vibration isolation beat a wall that low frequency ignores. invc.com
For residual low-frequency tones, consider duct ANC or resonator silencers — verify the vendor's band-limited claims on site. purilian.com
Verify with octave-band measurements at the property line, at night, across seasons — and publish the results.

Verdict. No — data-center noise is not a solved problem at the portfolio or community level. The mitigation toolbox is mature and a well-sited new facility can usually meet a defensible target, but the documented record is one of complaint-driven retrofit: sites built too close to homes, dBA ordinances that pass an intrusive hum, and barriers that fail at low frequency. The frontier is source elimination — liquid cooling, slower fans, magnetic bearings — combined with frequency-aware monitoring and enforceable performance, not ever-taller sound walls.

Method: built from three refined evidence sets — 10 normalized noise-control solutions with acoustic evidence and sources, 39 extracted records of community disputes across seven distinct facilities (duplicates collapsed into six representative case cards plus one finding), and 54 official planning and measurement requirements (several per document; not a count of jurisdictions). Live-web research performed 2026-08-29; searches returned 381 unique results on engineering practices, 1,283 on disputes, 2,127 on advanced technologies, and 1,879 on policy — candidate counts, not incidence statistics. dB figures measure sound pressure as reported by each source; vendor attenuation claims are not independently validated. Allegations in pending lawsuits are claims, not adjudicated facts.