Why we're opposed
The case against a data center in Henry County
Four documented impacts, and two areas where the evidence is still building. We label which is which, because a page that asserts everything with equal confidence deserves to be doubted.
How to read this page
Backed by government reports, measured data, or peer-reviewed research. You can take these to a Board meeting and defend them.
Real research and firsthand resident accounts, but the evidence base is thinner or still accumulating. Presented as what researchers and residents report — not as settled fact.
Our sources page lists everything, including a state study that argues data center tax breaks are good for Virginia. We'd rather you see that from us than discover we hid it.
Water
DocumentedCooling a data center consumes water, and much of it is lost to evaporation rather than returned. This was already identified as Henry County's single biggest concern before the July hearing.
50+
Cities with moratoriums
Localities that have paused data center development, many over water. [source]
Roughly 40% of US data centers sit in areas already under water stress. Where a facility draws from a public system, its consumption can translate directly into residential use restrictions during a drought — the resident pays for the shortfall in the form of a watering ban while the facility keeps cooling.
The risk is not only volume. University of Georgia researchers document effects on both surface water and groundwater, which matters enormously in a county where a large number of households are on private wells. A well that runs dry or turns is not a line item on a utility bill; it is a house that becomes unlivable.
This is why we ask for closed-loop cooling and no public water access — see the rules we want.
Power bills and the grid
DocumentedThis is the best-documented harm, and it does not come from an advocacy group. It comes from JLARC — the Virginia General Assembly's own nonpartisan research arm.
JLARC, December 2024
A typical Dominion residential customer could pay an estimated $14 to $37 more per month by 2040 — in today's dollars, before inflation.
Statewide, unconstrained data center growth is projected to raise generation and transmission costs by as much as $18 billion by 2040. About 340 data center buildings already require roughly 5 gigawatts of capacity, and that demand is expected to double within 15 years.
Read the JLARC reportThe mechanism is worth understanding, because it is the part industry representatives tend to skip. JLARC found that current rates do allocate costs appropriately to the customers who cause them — including data centers. The problem is what happens next: a large new customer forces construction of generation and transmission that would not otherwise be built, and those fixed costs enter the system that every customer pays into.
In other words, you do not need anyone to behave badly for your bill to go up. It goes up because the grid got bigger to serve someone else.
This is why we ask that developers fund their own substations and upgrades rather than socialize them — see the rules we want.
Noise
DocumentedData centers run every hour of every day. The cooling equipment that makes that possible produces a continuous industrial drone that neighbors in Northern Virginia describe as inescapable.
What 40–59 decibels actually sounds like
Noise measured on residential property near Northern Virginia data centers, placed against familiar sounds. The EPA identifies this range as loud enough to interfere with sleep, work, and exercise — and unlike a vacuum cleaner, it never switches off.
Measurements reported by the Environmental and Energy Study Institute. See source.
Chandler, Arizona is the clearest cautionary example. A data center moved in beside the Brittany Heights neighborhood in 2014; residents reported a hum that never stopped, even overnight. After nearly a decade of complaints the city amended its zoning code in 2022 to make siting a data center harder, and in 2025 the council unanimously rejected a new one — with noise central to the opposition.
The lesson Henry County should take is about sequence. Chandler got its rules after living with the problem. We still have the chance to get them first.
This is why we ask for a binding noise limit measured at the property line — including low-frequency content, which a simple dBA cap misses entirely. See the rules we want.
Low-frequency noise and health
Emerging evidenceResidents near data centers increasingly report a category of sound that a standard decibel meter does not capture — felt more than heard. The research here is real but still accumulating, so we present it as what researchers and residents report.
What the research currently shows
- A peer-reviewed systematic review in Science of the Total Environment found associations between low-frequency noise and infrasound exposure and self-reported annoyance, sleep problems, difficulty concentrating, and headaches — while explicitly noting the evidence base remains limited. [source]
- In one survey of people reporting infrasound exposure, insomnia and concentration problems were reported by roughly two-thirds of respondents. [source]
- Research on low-frequency noise and sleep finds effects on sleep quality, time taken to fall asleep, and morning tiredness. [source]
- Complaints about data center sound that does not register on standard meters are a growing and documented category near AI facilities. [source]
We want to be careful here, because this is where opponents of regulation will look for a weak claim to attack. We are not saying that a data center will make people sick. We are saying something narrower and much harder to argue with:
There is enough credible research, and enough consistent testimony from people already living beside these facilities, that a county writing rules from scratch should measure low-frequency noise and set a limit — rather than write a dBA cap that by construction cannot detect the thing residents are complaining about.
That is a reasonable ask regardless of where the science eventually lands. If the effect turns out to be small, a low-frequency limit costs the developer very little. If it turns out to be real, the limit is the only thing standing between a resident and a problem with no remedy.
Wildlife, pollinators, and livestock
Emerging evidenceResidents raised habitat and animal effects at the July hearing. The general science on human-made noise and animals is well established; the specific application to data centers is not yet studied directly, which is precisely why we ask for a study.
What is well established
A meta-analysis published by the Royal Society found that anthropogenic noise produces measurable physiological stress responses and behavioral change across animal taxa, with documented effects on communication, reproductive success, and foraging. Chronic noise — the 24/7 kind, not the occasional kind — is the specific case researchers flag as most damaging.
What is newer and less settled
Insects communicate through substrate vibration largely below 1000 Hz — overlapping directly with the frequency range of industrial vibration. A 2024 experiment found noise reduced bumblebee-mediated pollination success in tomatoes, and separate work measures how railway vibration propagates into beehives. Both are preprints and should be read as early findings, not conclusions.
Henry County sits in a watershed with federally listed species, including the Roanoke logperch and the James spinymussel. That is not a rhetorical flourish — it is a legal lever. Federal protections attach to those species regardless of what the state or the county would prefer, which makes an endangered species study one of the most durable requirements a locality can attach to a permit.
This is why we ask for a multi-year environmental and species study before approvals proceed — see the rules we want.
Knowing this is only useful if your supervisor hears it.
The Planning Commission is writing the replacement rules now. Pick the argument that matters most to you, put it in your own words, and send it to the person who represents your district.