Kasey Fowler-Finn's data center noise guidance recommends 1,500-foot setbacks over 4.5-foot walls

The wall that would quiet a data center's hum is four and a half feet thick. For the 63-hertz tone large cooling and HVAC systems emit, Kasey Fowler-Finn, a biology professor at Saint Louis University, estimates that thickness; down near 20 hertz the estimate rises to 14 feet, which her guidance describes as not remotely practical for a building enclosure — and a four-inch wall already handles the roughly 1,000-hertz whine of a vacuum cleaner.

Fowler-Finn released model data center noise guidance last month in coverage published by Construction Dive, based on research she says she has been conducting since 2001. Her subject is the low-frequency hum that large cooling and HVAC systems emit, which she argues travels further and registers more than the high-frequency noise many local ordinances are written to catch. She told Saint Louis Public Radio on Sept. 24 that hums concentrated in single frequencies stand out from background noise "no matter how loud or what frequency that background noise is in."

Construction fixes run into the physics. Sound loses intensity as it spreads — 6 decibels each time the distance from the source doubles, and 30 decibels by the time it has covered 3,000 feet, she says — and it loses more as structures absorb it, but absorption does less work at low frequency because the wavelengths are long. Low frequencies pass through walls and windows with little loss, since ordinary building materials are ineffective at attenuating them, and they can generate secondary noise by inducing vibrations and rattles in nearby structures.

What an ordinance measures and what a neighbor hears are different things. A facility can be compliant on paper while still delivering daytime and nighttime noise exposure that Fowler-Finn links to numerous health concerns — exposure that at very low frequencies is inaudible and still, in her term, physiologically relevant. Her derived wall thicknesses run the length of the spectrum:

Commonly used A-weighted and C-weighted decibel metrics are adequate as a starting point, she argues, but they need to be paired with third-octave band sound pressure testing covering 20 to 200 hertz, which makes any tonal or narrow-band component of a facility's noise independently identifiable rather than averaged away.

With walls impractical, the guidance puts the recommended setback at a minimum of 1,500 feet, which converts an acoustics question into a site-selection question. A developer shopping for a parcel is also shopping for the buffer that comes attached to it: fifteen hundred feet on every side is not a distance added later but land the project holds, competing with the substation, the yard and the access road for the same acreage. The setback also sits inside the 3,000-foot span over which she calculates a 30-decibel loss, so the buffer appears meant to work alongside measurement and enclosure rather than instead of them.

Site plans are where these numbers become binding, and the setback is the most writable part of the guidance: it is a figure that appears on a drawing. The testing recommendation is where the leverage likely sits, though, for a reason that has nothing to do with acoustics — it changes what a project has to prove. Local conditions accumulate: a sound study required in one jurisdiction becomes the reference point a neighboring one is handed. Model guidance is built to be borrowed, which is a different thing from a rule; none of it binds anybody until a planning board or a county commission says it does.

PWD has argued that consent, rather than capital, is the scarcest input in data-center development, and noise is one of the few consent problems that responds to engineering — provided it is priced at the parcel rather than after the first complaint. The guidance is evidence for both halves of that argument: the engineering is legible, and the currency it demands is land.

The coverage leaves the pricing open: no jurisdiction is identified as having adopted the guidance, no cost is attached to a 1,500-foot setback or to third-octave testing, and no developer response to either appears. Fowler-Finn's research program, by her account, stretches back to 2001; the model guidance is the new object in the file.

The 20-to-200-hertz band is the piece most likely to move. An ordinance that writes that testing into its conditions would push a research finding into the siting file, where the buffers have already been bought or lost.