“A data center automatically raises every household’s electric bill.”Common opponent claim
Rate design
−6% vs. +2.1%conflicting historical causal estimates; 2015–24 / 2010–24
EPRI paper
Quick reference
MEETING COPY · 07/14/26
Claim vs. evidence report card
Grades describe the claim as written—not whether a project is “good” or “bad.” Every number is dated and linked.
“Existing customers have never paid data-center connection costs.”Common developer / utility claim
Unsupported
$4.3BPJM connection projects approved in 2024; advocacy analysis
UCS analysis
“One facility brings hundreds of permanent local jobs.”Common booster claim
Unsupported
≈50typical full-time workers at a 250,000-ft² Virginia facility
Virginia JLARC
“Construction creates a large, real—but temporary—jobs surge.”Common booster claim
Supported
≈1,500 peak12–18 months per building; Virginia industry interviews
Virginia JLARC
“Every data center consumes the same huge amount of municipal water.”Common opponent claim
Mixed
18k–550k gal/dayon-site range; cooling and climate drive the spread
Ceres analysis
“If sound meets the A-weighted limit, neighbors cannot be affected.”Common developer claim
Unsupported
40–59 dBAcomplaint-site measurements; constancy and low frequency matter
Virginia JLARC
“A data center always produces a huge net local tax windfall.”Common booster claim
Mixed
$0.4M–$10.8M5-year tax range on the same $150M equipment example
Virginia JLARC
“Tax incentives are unusual exceptions.”Common opponent claim
Unsupported
38 statesdedicated data-center incentives as of Apr 2026
NCSL tracker
“Nearby data centers have been proven to collapse home values.”Common opponent claim
Unsupported
No clear evidenceVirginia review found no observed effect yet; not proof of no effect
Virginia JLARC
“One facility inevitably causes local blackouts.”Common opponent claim
Unsupported
Risk, not destinyinterconnection study, firm supply and curtailment terms decide exposure
Virginia JLARC
S · Supported: claim matches the cited evidence.
M · Mixed: true only in part or varies materially.
R · Rate design: the tariff and regulator decide who pays.
U · Unsupported: cited evidence contradicts it or does not establish it.
Fundamentals
The community relationship is a lifecycle
A project changes shape over time. Construction benefits cannot be presented as permanent operations benefits, and operating obligations must survive ownership changes.
Land & zoning
Setbacks, road access, buffers, building height, generators and future phases are decided before excavation.
Utility study
The utility models new substations, transmission, generation, security deposits and the project’s requested MW.
Water plan
Source, peak-day demand, cooling technology, drought response and wastewater capacity determine local exposure.
Construction
A building commonly takes 12–18 months; a campus can build in phases for five years or more.
Operations
Continuous load, cooling, fans, backup testing, equipment taxes and a relatively small permanent staff remain.
Refresh or exit
Servers may refresh roughly every few years; a customer exit can strand dedicated utility assets without contract protection.
The honest one-line model: the facility creates a large, steady industrial load. Whether that load lowers unit costs by spreading fixed costs—or raises other customers’ bills by triggering scarce new infrastructure—depends on timing, grid headroom, contract duration, and who the tariff assigns the risk to.
Working model
Construction and operations are different ledgers
Demand a two-column answer whenever a presentation says “economic impact.” The peak phase and the durable phase should never be blended.
| Domain | Construction phase | Operating phase | Document to demand |
|---|---|---|---|
| Electricity | Temporary construction service; modest beside final load. | Large, continuous MW demand plus dedicated or shared grid assets. | Interconnection study, service agreement, tariff and regulator order. |
| Jobs | Up to ≈1,500 workers at peak for a Virginia facility example; trades rotate by phase. | ≈50 full-time workers for a typical 250,000-ft² Virginia facility; about half contracted. | Annual headcount by worksite, residency, employee/contractor status and wage band. |
| Water | Dust control, concrete and worker use; usually not the lifetime driver. | Cooling can range from near-dry to hundreds of thousands of gallons/day; peak heat days matter. | Monthly and peak-day withdrawal/consumption, source, WUE, drought stages and discharge. |
| Noise | Trucks, grading and building work; time-limited but campus phasing can extend years. | Fans, chillers and transformers create continuous tonal/low-frequency sound; generator tests add events. | Preconstruction model plus post-opening dBA, dBC and one-third-octave measurements at homes. |
| Land & homes | Clearing, road wear, views and traffic are immediate. | Industrial massing and infrastructure remain; current Virginia evidence does not establish a price effect. | Viewshed, setbacks, landscaping bond, traffic plan and repeat-sales appraisal method. |
| Taxes | Permit, sales, lodging and worker-spending effects may spike. | Real and equipment-property taxes can be substantial, but exemptions, rebates and depreciation change net value. | Twenty-year gross-to-net fiscal model with every exemption, rebate and service cost. |
Six claim domains
What the evidence actually lets you say
National evidence helps choose questions. Only the project tariff, permits, contracts and measurements can answer the local version.
Electricity bills
Verdict · mixed; rate design is decisive
Two June 2026 causal working papers disagree. An EPRI-authored paper estimates a doubling of data-center capacity lowered residential rates about 4%, with 2019–24 growth associated with a 6% decrease. An MIT CEEPR paper estimates entry increased residential prices 2.1% during 2010–24, with larger increases at investor-owned utilities.
- Concrete bill translation: 2.1% on a $150 monthly bill is $3.15—not a forecast for your town, just a scale conversion of the MIT estimate.
- Forward risk: UCS, an advocacy nonprofit, reports $4.3 billion of 2024 PJM connection projects were broadly allocated rather than directly assigned.
- Do not infer: today’s rate increase is not automatically caused by one facility; fuel, storms, existing rate cases and regional capacity markets also move bills.
Jobs & payroll
Verdict · construction strong, operations lean
Virginia’s official 2024 audit reports that a 250,000-ft² center typically has about 50 full-time workers, roughly half contractors, while peak construction can put about 1,500 workers on site for a 12–18 month building cycle.
- Ask for annualized jobs: one worker employed for one year equals one job-year; “1,500 jobs” at peak is not 1,500 permanent positions.
- Ask who is local: payroll impact depends on workers’ home ZIP codes and whether vendors are local.
- Do not count twice: distinguish direct on-site staff, contracted maintenance, construction trades and modeled indirect jobs.
Water
Verdict · site and cooling design decide
Ceres, a sustainability advocacy nonprofit, gives an on-site range of 18,000–550,000 gallons/day depending on size. DOE says facilities can reach millions of gallons/day, while advanced cooling can reduce needs by more than 90% in some designs.
- Withdrawn ≠ consumed: withdrawals enter the site; consumption is not returned to the immediate watershed, commonly through evaporation.
- Peak beats average: request the hottest-day number and drought-stage behavior, not only an annual average.
- Include indirect water: electricity generation may use water outside the county; keep it separate from the local utility ledger.
Noise
Verdict · audibility and harm are not the same test
Virginia complaint-site measurements ranged from 40–59 dBA. JLARC found the constant hum can affect well-being even when it stays below common 55–60 dBA ordinance limits; A-weighting also discounts low frequencies.
- Measure the receptor: require modeled and measured sound at the nearest home, not only at equipment or property lines.
- Specify metrics: include dBA, dBC, tonal penalties, nighttime limits and one-third-octave data.
- Require a cure: name response deadlines, independent testing, retrofit triggers and enforceable penalties.
Taxes & abatements
Verdict · potentially substantial, never quote gross alone
JLARC’s same-facility example—$150 million of computer equipment—produced a five-year local tax range of $0.4 million to $10.8 million under different Virginia rates and depreciation. As of April 2026, NCSL reports 38 states offer dedicated incentives.
- Net means net: subtract sales-tax exemptions, property abatements, rebates, infrastructure subsidies and new service costs.
- Model replacement: equipment depreciation reduces the base, while periodic server refreshes may reset it.
- Protect volatility: avoid pledging recurring services against a single customer without a reserve and downside case.
Property, land & reliability
Verdict · local harms plausible; broad price claims unproven
JLARC found no evidence yet of lower nearby home values in its Virginia review, but also found 29% of operational properties were within 200 feet of residentially zoned land and about 10% of sites appeared to generate problematic noise.
- No evidence ≠ no effect: the review relied on interviews and a tight housing market, not a national repeat-sales causal study.
- Reliability is conditional: reserve margins, deliverable generation, transmission, backup systems and curtailable-load terms matter.
- Site planning works: industrial zoning, setbacks, buffers and sound modeling reduce conflict more reliably than complaint response after opening.
Edge & advanced
The six protections that change outcomes
Promises are not protections. A protection has a number, a responsible party, an independent verification method, a remedy and a survival clause.
Separate large-load tariff
Assign incremental service costs to the project class. Demand minimum bills, demand ratchets, collateral and contract terms long enough to cover dedicated assets.
Beneficiary-pays interconnection
Identify every substation, feeder and transmission upgrade; specify direct assignment, shared-allocation formula, refunds and what happens if the project cancels.
Exit & underuse protection
Use take-or-pay minimums, credit support, security deposits and early-termination payments so unused infrastructure does not migrate to other ratepayers.
Measurable community agreement
Put local hiring, road repair, sound limits, water reporting and emergency coordination into an enforceable agreement with dates—not a voluntary benefits brochure.
Disclosure without NDA blind spots
Publish load by phase, tax terms, water demand, acoustics and public obligations. Redact genuine trade secrets narrowly; do not hide public costs or performance.
Independent audit & remedy
Let the town select the auditor at developer expense. Define reporting cadence, complaint response, cure deadlines, penalties, bond draws and successor-owner duties.
Legislative temperature: NCSL’s July 1, 2026 tracker lists proposals in 15 states to ban, pause or condition data-center development; statuses range from introduced to failed, vetoed or passed by a legislature. This is a fast-changing signal of scrutiny—not evidence that a statewide moratorium is in force where you live.
Bring this to the meeting
Nine questions the record should answer
0 of 9 documented
What is the peak MW by phase, load factor, customer class and tariff?
Why it mattersMarketing square footage does not size grid exposure; MW, ramp schedule and contractual demand do.
A good answer“Phase 1 is 96 MW in 2028, 92% expected load factor, served under tariff LL-2 filed in docket 26-104.”
Which grid upgrades are triggered, what does each cost, and who pays?
Why it matters“We pay our fair share” is not an allocation method; regional and local upgrades can be treated differently.
A good answerA line-item study naming direct-assigned and shared costs, refund rules, contingencies and the regulator-approved allocation.
What protects other customers if load arrives late, underperforms or exits?
Why it mattersUtility assets outlive a short service contract; forecast error can become stranded cost.
A good answerMinimum monthly bill, demand ratchet, collateral, parent guarantee and early-termination payment matched to asset life.
What are average and peak-day water withdrawal and consumption?
Why it mattersAnnual averages hide hot-day demand; withdrawal and consumption are not interchangeable.
A good answerMonthly gallons by source, hottest-day gallons, consumed versus discharged volume, WUE, reclaimed share and metering plan.
What changes during each drought stage, and who has priority?
Why it mattersA low annual average can still collide with residential and agricultural demand during a dry, hot week.
A good answerBinding stage triggers, curtailment percentages, alternate source, no potable makeup at specified stages and public reporting.
Where, when and with which frequency metrics will sound be tested?
Why it mattersdBA alone can understate low-frequency tonal hum; daytime averages can hide nighttime exposure.
A good answerNearest-home receptors, day/night dBA and dBC, one-third-octave bands, generator testing, independent commissioning and retrofit trigger.
What is the 20-year net public revenue after every incentive and cost?
Why it mattersGross assessed tax is not net revenue; depreciation, rebates, roads, utility work and emergency services change the ledger.
A good answerYear-by-year base, rates, depreciation, refresh assumptions, abatements, rebates, public costs and low/base/high scenarios.
How many permanent on-site jobs—not job-years—will exist after buildout?
Why it mattersPeak construction, annual job-years, contractors and permanent employees are routinely blended.
A good answerAnnual worksite headcount by phase, occupation, employee/contractor, wage band, benefits and worker home ZIP code.
Who verifies each promise, and what happens when it is missed?
Why it mattersA target without data access, an auditor, a deadline and a remedy is public relations, not risk control.
A good answerNamed independent auditor, public annual report, complaint channel, cure period, escalating penalty or bond, and successor-owner clause.
Common mistakes
Six ways a meeting goes off the rails
Good evidence can still produce a bad decision when units, time periods or comparison groups are mixed.
Blaming one project for the whole bill
Separate fuel, capacity, transmission, distribution, storm recovery and project-specific upgrades. Regional costs do not stop at county lines.
Calling peak workers permanent jobs
Report peak headcount, annual job-years and steady-state on-site positions as three different numbers.
Quoting gross taxes as net benefit
Model exemptions, abatements, depreciation, rebates, infrastructure, services and downside—not just the first-year assessment.
Treating gallons as one metric
Keep withdrawal, consumption, discharge, peak-day demand, potable/reclaimed source and indirect power-sector water separate.
Using ordinance compliance as a nuisance test
A-weighted limits target loudness, not every persistent low-frequency or tonal effect. Require the right metrics and a post-opening cure.
Performing “both-sides” neutrality
Neutral process does not mean equal grades. A claim contradicted by the cited evidence should be labeled unsupported, whoever makes it.
Source register
What was verified, and by whom
“Public” means government or legislative research. Advocacy and industry-funded evidence is useful but labeled so readers can weight incentives.
- Gallup, Americans Oppose AI Data Centers in Their Area (fielded Mar 2–18, published May 13, 2026): 71% opposed locally; 48% strongly. Pollster
- MIT CEEPR, Who Pays for Growth? (Jun 2026): facility entry and retail price estimates through 2024. Academic working paper
- Watten, Bistline & Blanford, Have Data Centers Raised Your Electric Bill? (Jun 2026): contrary causal rate estimate and future-constraint caveat. EPRI authors · working paper
- Virginia JLARC, Data Centers in Virginia (Dec 2024; accessed Jul 2026): jobs, taxes, water, siting, noise, property and grid risk. State audit
- U.S. DOE, Data Centers for Tribal Economic Development FAQ (May 2026): water variability and cooling reduction potential. Federal public
- NCSL, Subsidizing Servers (updated Apr 1, 2026): 50-state tax-incentive landscape. Legislative public
- NCSL, Which States Are Banning Data Centers? (updated Jul 1, 2026): 15-state proposal/status tracker. Legislative public
- Union of Concerned Scientists, PJM connection-cost analysis (Sep 2025): $4.3B claim and allocation method. Advocacy nonprofit
- Ceres, Drained by Data (2025): direct/indirect water framing and 18,000–550,000 gal/day range. Advocacy nonprofit