What Are the Main Types of Data Center Noise?

Noise is the complaint local officials hear most often once a data center opens, and it is the impact many ordinances still handle worst. The reason is that a data center does not make noise the way a factory or a highway does. Its sound is continuous, it runs for years without pause, and much of its energy sits at low frequencies that ordinary noise limits are built to discount. Knowing the main types of data-center noise — and the ways to measure them — is the difference between a rule that looks strict on paper and one that actually protects the homes nearby.

The two questions a noise rule has to answer

Any honest look at data-center noise starts with two separate questions. The first is how loud the sound is. The second is how deep it is — how much of its energy sits in the low frequencies. A standard decibel meter answers the first question well and the second one badly. That gap is where most of the conflict between data centers and their neighbors lives: a facility can satisfy a loudness limit at the property line and still send a low, steady drone into bedrooms at night.

The main sources, and why the sound never stops

Data-center noise comes from the machinery that keeps thousands of servers cool and powered. Four sources account for most of it:

Chillers and cooling plants — the largest and most constant source, running whenever the servers run, which is essentially all the time.
Cooling fans — large arrays moving air across equipment, adding a steady broadband whir.
Transformers and electrical gear — a continuous electrical hum at the site’s high-voltage connection.
Backup generators — intense but intermittent, firing during outages and during scheduled testing, which is why ordinances often restrict testing hours.

Together these produce a sound that is steady and concentrated at low frequencies — closer to a held bass note than to the rising and falling noise of traffic. A cooling plant is that bass line, held without pause for years.

Why low-frequency noise behaves differently

Low notes pass through walls and windows that stop higher ones. Think of the bass from a car idling at a stoplight: you do not hear the song, you feel the thump, because the low frequencies travel through steel and glass while the high ones do not. That is why residents can describe a serious problem inside their homes while an outdoor loudness reading looks unremarkable. A 2003 review for the U.K. Department for Environment, Food and Rural Affairs found that low-frequency noise can cause “extreme distress” to the people sensitive to it, and concluded that the claim such noise has “ruined” lives “is not an exaggeration.”

How data-center noise is measured

There are three ways to capture the sound itself, and they do not measure the same thing:

A-weighted decibels (dBA) — the standard measure of loudness, weighted to match human hearing, which means it discounts exactly the low-frequency energy a data center produces most. The World Health Organization’s guidance, restated in the DEFRA review, warns that “A-weighting underestimates the sound pressure level of noise with low frequency components.” A facility can meet a 55-dBA property-line limit and still generate a persistent low-frequency drone.
C-weighted decibels (dBC) — flatter across the low frequencies, so it captures energy that an A-weighted reading minimizes. WHO guidance quoted in the review concludes that “a better assessment of health effects would be to use C-weighting.” It is not a complete fix on its own, but it exposes what dBA hides.
Low-frequency and infrasound analysis — measuring the specific low and sub-audible bands directly, rather than collapsing everything into a single number.

The weighting is only half of it. A technically serious standard also sets the conditions under which the sound is measured:

A baseline — the ambient sound level before the facility is built, so its own contribution can be isolated.
Modeled sources and receptor locations — where the sound originates and where it is measured, typically at the nearest homes rather than at the fence line.
An averaging period, with separate day and night limits.
Tonal penalties — a stricter allowance for steady, single-pitch sounds, which are more annoying than their raw decibel level suggests.
Post-construction verification and continuing monitoring — so the rule is tested against the operating facility, not just the modeled permit, and the operator knows what happens if the finished building fails.

Why one decibel number — or a big setback — is not enough

Most local rules name noise, attach a single number, and lean on distance. But distance is not an acoustic outcome. Topography, equipment orientation, weather, and multiple sound sources can make a modeled buffer perform differently in practice. In its review of local data-center ordinances, American Buildout found that nearly three-quarters addressed noise, yet only about a third set a numeric standard, and just 14 percent addressed C-weighting, infrasound, or low-frequency sound at all. A large setback can reduce exposure, but it does not prove that a compliant facility will meet a measured low-frequency standard at someone’s home. Acoustic limits depend on local conditions and should be set with a qualified acoustical engineer; the point here is what to measure, not a single number to copy.

Sources

Geoff Leventhall, Peter Pelmear, and Stephen Benton, “A Review of Published Research on Low Frequency Noise and its Effects” (report for the U.K. Department for Environment, Food and Rural Affairs, 2003).

World Health Organization, “Guidelines for Community Noise” (1999).