No study measures health outcomes near data-centers, so noise limits rest on dose-response borrowed from traffic

Why this is its own facet

The parent issue concerns emissions and measurement. This one concerns a gap in the evidence chain that both sides of every local dispute currently paper over: nobody has measured what continuous industrial tonal noise does to the people living next to it.

What the existing evidence actually covers

The epidemiology invoked in these disputes is robust but it is transport epidemiology. Münzel and colleagues report a pooled coronary heart disease risk of 1.08 (95% CI 1.04–1.13) per 10 dB(A) increase in road traffic noise, a 14% increase in stroke risk per 10 dB(A) in a Danish cohort, an odds ratio of 1.14 for hypertension per 10 dB(A) of night-time aircraft noise from the HYENA study (~5,000 participants), and an estimated 1,685,000 DALYs lost annually in Western Europe from traffic noise (Münzel et al., 2018). WHO Europe night noise guidance rests on the same body of work (WHO Europe).

Why the transfer is an assumption rather than a finding

Road, rail, and aircraft noise is intermittent, broadband, and varies across the day. data-center noise is continuous, spectrally narrow, and concentrated at low frequencies — differing in the two variables the mechanism is thought to depend on: arousal response to noise events, and frequency content that penetrates building envelopes. The direction of that difference is not obvious. Continuous exposure could be worse (no recovery periods, no habituation to a tone) or better (no startle events). Neither has been measured.

This matters because regulators are setting numeric limits now. Prince William County's proposed residential limits of 48 dBA day / 43 dBA night, and its proposed octave-band criteria, are reasonable engineering judgements that cannot currently cite a study of this source type (Prince William County, 2025). Operators can point out that health claims are extrapolated; residents can point out that absence of study is not absence of harm. Both are correct, which is why disputes do not resolve on evidence.

Current state vs. desired state

Current: zero peer-reviewed studies measuring sleep, blood pressure, or cardiovascular endpoints in populations adjacent to operating data-centers, despite hundreds of facilities having operated near housing for a decade or more.

Desired: at least one measured cohort with characterised exposure, so that limits and setbacks can be grounded in evidence specific to this source type.

Who is affected

Residents, whose complaints are treated as subjective; operators, who face limits they can argue are unfounded; and regulators, who must set numbers without a study to cite.

Impact if unaddressed

Limits will continue to be set by political negotiation and litigation rather than dose-response, producing both under-protection in permissive jurisdictions and unnecessary cost in strict ones. Many facilities now have a decade of operation next to stable residential populations — a natural experiment whose baseline becomes harder to reconstruct as clusters densify.