Of everything communities raise about data centers, noise is where the case is strongest on the merits and weakest on evidence quality. It is also the most common post-approval grievance — the thing that shows up eighteen months after the ribbon-cutting, when nobody has any leverage left.
And it produces a pattern that looks, at first, like people making things up.
A facility is tested. It passes. Residents keep complaining. The operator, sincerely, points at the compliance report. The residents, sincerely, cannot sleep. Everyone concludes the other side is lying.
Nobody is lying. The ordinance is measuring the wrong thing.
Three ways standard noise rules miss
A-weighting systematically understates data center noise. The dBA scale — the one in nearly every local ordinance — deliberately de-emphasizes energy below roughly 200 Hz, because human hearing is less sensitive there. That’s a reasonable choice for traffic or a nightclub. It is exactly wrong here, because transformer hum and fan blade-pass frequencies sit precisely in the de-emphasized range. A facility can pass a 55 dBA property-line test and still produce the specific complaint pattern residents describe.
Tonal noise generates complaints far below broadband thresholds. A steady hum at an identifiable pitch is dramatically more annoying than the same energy spread across a spectrum. Ordinances that set one broadband limit and stop don’t capture this at all.
Duration drives complaints, not peak. A loud event that ends is tolerable. A moderate sound that never stops is what breaks people. Almost every ordinance is written around peaks.
Put those together and you get the finding that should reframe the entire conversation. An independent audit in one large market concluded that data center noise is “low-frequency noise that rarely violates ordinances,” with roughly a third of facilities in that state sited near residential areas.
Compliance and complaint are decoupled. That is not an accusation against any operator. It’s a diagnosis of the measuring instrument.
The physics that makes this negotiable
Here’s why noise is such productive ground: it’s quantifiable in a way most siting disputes are not.
Cooling fans run roughly 70 to 85 dBA at fifty feet. Generator load tests run 80 to 100 dBA at the property line. Attenuation for a point source is about 6 dB per doubling of distance.
Which means 70 dBA at fifty feet is roughly 50 dBA at eight hundred feet, and about 40 at half a mile, over flat open ground.
That is the entire setback argument, in arithmetic anyone can check. Not a values claim. Not a vibe. A calculation, with an answer.
And mitigation has published numbers too, so the conversation can be engineering rather than assurance:
- Acoustic louvers and shrouds: 5–15 dBA
- Residential-grade silencers: 10–20 dBA
- Sound walls and berms: 10–20 dBA, over a limited angle only
- Increased setback: cheapest and most reliable of all
- Vegetative buffers: negligible
That last one deserves a flag. Trees do very little for noise. When a site plan presents a landscape buffer as acoustic mitigation, that’s a tell — either the applicant doesn’t know, or is hoping you don’t.
What belongs in the permit
If your ordinance sets one dBA number, you have a standard that a facility can satisfy while generating a complaint file. Fix the instrument:
- Property-line limits in dB(A) and dB(C) — the C-weighted scale keeps the low-frequency energy the A scale discards.
- A tonal penalty, so an identifiable hum is treated as what it is.
- Nighttime limits separately stated.
- Real baseline measurements taken before construction, so there’s something to compare against. Rural nighttime ambient is far lower than people assume, which is why a limit that sounds generous can be a major change.
- Mandatory post-construction Type 1 verification at the applicant’s expense, with tonal and low-frequency analysis.
- Penalties large enough to matter, and a complaint process with deadlines attached.
One more, and it’s the important one: noise conditions belong in the permit, not in a later enforcement action. A condition attached to a conditional use permit stays live and revocable for the life of the facility. A promise in an agreement requires you to sue.
The honest limits
I’m not going to overclaim here, because the overclaim is how communities lose this argument.
Exposure is documented. The causal health literature specific to this noise — chronic, low-frequency, industrial, residential-adjacent — is thin. Neither “proven harm” nor “no evidence of harm” is an accurate statement of where the science currently sits.
And the study quality in this space has been poor on both sides. In the largest market, nobody has yet produced a Type 1 measurement campaign with proper tonal and low-frequency analysis. That absence is itself the most useful fact available, because it means the first ask in any noise dispute is obvious: let’s go measure it properly, together, on terms we agree to in advance.
The eighteen-month lesson
One operator spent eighteen months under sustained public pressure before retrofitting fan shrouds — equipment it could have specified at design stage for a fraction of the eventual cost, and with none of the reputational damage.
That’s the whole argument for getting this right at the permit rather than in year two. The shrouds were always going to get installed. The only variable was whether they were installed cheaply, quietly, as part of the original design, or expensively, publicly, after the local paper had run four stories about people who couldn’t sleep.
Every condition you want operating in year fifteen has to be written in year one, with a verification mechanism and a consequence attached.
Noise is where that lesson gets learned the hard way, over and over, in county after county — because a facility passed a test that was never designed to detect the thing people were complaining about.