My thesisUpdated Aug. 24, 2026

Local communities have never had more power to demand the best from data centers.

If a data center is coming to your area, say: “This is what I want. These are my standards.” Put them in writing before the project is approved.

If local government will not insist on those terms, elect one that will. We have never had a better shot at good jobs and at building the future of intelligence in America. The examples below show what a good deal can include.

Start with the standards

01

Get it in writing before you say yes

A promise in a press release is not a standard. Put the cost, limit, measurement, and penalty in the permit, tariff, tax agreement, or public contract.

01

Power bills

The data center pays for the grid work it causes, signs a long-term minimum-payment contract, and covers the risk if its forecast load never arrives.

02

Water

The public sees annual and peak-day use, the water source, drought rules, and the cooling design before the vote.

03

Homes

Industrial zoning, real setbacks, sound modeling, low-frequency noise limits, generator hours, and penalties are written into the permit.

04

Jobs

The agreement names wage, apprenticeship, local hiring, and reporting requirements. Construction claims and permanent jobs are reported separately.

05

Taxes

The public sees the tax schedule, exemptions, infrastructure costs, and clawbacks before officials approve the deal.

06

Roads and services

The project pays for road damage, truck management, fire and emergency planning, and any added public capacity it needs.

07

Grid reliability

The utility can delay energization until the system is ready. The project brings supply, storage, phased demand, or limited curtailment where the grid needs it.

08

Enforcement

Water, power, jobs, taxes, noise, and emissions are reported publicly. Missed promises trigger a defined remedy.

02

The questions people are asking

These are fair objections. Here are specific answers, the sources behind them, and the terms a community can demand.

Ohio and Virginia

These states show why blanket approval is a mistake and how public pressure can change the rules.

“This is already terrible in Ohio. Why are you defending it?”

I am not defending every Ohio project: Ohio’s consumer advocate says a hyperscale data center can use as much electricity as 100,000 homes, and new transmission and substations can leave other customers holding part of the bill if the utility contract is weak.

Ohio also shows that residents can force a better rule. In 2025, the Public Utilities Commission approved an AEP Ohio tariff that requires qualifying new data centers to pay for at least 85% of contracted capacity for up to 12 years, even when they use less. That does not settle water, land, noise, or tax questions. It does put more of the power-system risk on the customer that caused it.

Put in writing: Apply the same cost protection outside AEP territory. Publish the water estimate, road plan, emergency-service plan, noise limits, and tax terms before rezoning.

“Virginia proves that data centers are a bad deal.”

Virginia proves that weak siting and utility rules have consequences. The state watchdog found data centers built too close to homes, low-frequency noise complaints, and a forecast in which a typical Dominion residential customer could pay $14 to $37 more per month for generation and transmission by 2040. That forecast assumes very large, unconstrained growth; it is not a bill increase that has already happened.

The same report found 74,000 jobs supported, $5.5 billion in labor income, $9.1 billion in state GDP, and large local tax receipts. Most of the employment benefit came from construction. Virginia’s record contains both the benefit and the failure to set terms early enough.

Virginia changed its law in 2026. For a proposed facility of 100 megawatts or more, a locality now must get a sound assessment near homes and schools. It may also require study of water, farmland, parks, historic sites, and forests. The utility must describe the substations and transmission voltage needed for the project.

Put in writing: Use the new assessment before approval. Keep industrial facilities away from homes and schools, and put cost allocation, sound limits, water use, and transmission routes in public documents.

Power and water

The power and water objections are real. National averages cannot tell you whether a specific project fits a specific town.

Does a data center use too much electricity?

The national load is large and growing fast: Lawrence Berkeley National Laboratory estimated that U.S. data centers used 176 terawatt-hours in 2023, or 4.4% of U.S. electricity. Its 2028 scenarios range from 325 to 580 terawatt-hours, or 6.7% to 12% of U.S. use.

That is a reason to require a power plan, not a reason to accept a vague promise that the grid will catch up. A project should show its phased demand, the generation and transmission needed to serve it, who pays, and what happens during the few hours when the system is tight.

Put in writing: Do not approve the full campus on the basis of a speculative load request; tie each phase to power that is funded and ready.

Will it raise my electric bill?

It can, because new generation, substations, and transmission have fixed costs. If the data-center load arrives late, uses less power than promised, or closes, those costs can move to other customers. Virginia’s watchdog modeled a $14 to $37 monthly increase by 2040 for a typical Dominion household under its high-growth case.

The answer depends on the tariff, which is the utility’s price and contract rule. Ohio’s AEP tariff requires large new data centers to keep paying for at least 85% of contracted capacity for up to 12 years. A town should ask for that kind of protection before supporting the project.

Put in writing: Require a separate large-load contract, customer-funded connection work, a long minimum-payment period, financial security, and an exit charge if the load disappears.

Will it take our water?

There is no honest answer without the cooling design and the local water balance. LBNL estimated 66 billion liters of direct U.S. data-center water consumption in 2023. It also estimated nearly 800 billion liters of indirect water consumption at the power plants that supplied the electricity.

The building-to-building range is enormous. In Virginia, most buildings in the state review used no more water than an average large office building, about 6.7 million gallons per year. Eleven buildings used more than 50 million gallons, and one used 243 million. Just over one-third of Virginia data-center water came from reclaimed water in 2023.

Quincy, Washington, uses a reuse system. Douglas County, Georgia, sends treated wastewater to Google. Compass reports that its Red Oak, Texas, campus uses no city water for cooling. Those are design choices a town can require or reward.

Put in writing: Publish annual and peak-day use by source, set drought rules, price drinking water correctly, and prefer reclaimed water, closed-loop systems, or air cooling where the full power-and-water tradeoff works locally.

Will it make the grid less reliable?

A very large load can make reliability worse if it connects before generation and transmission are ready. It can also help finance new supply and operate more flexibly than a town may assume. Duke researchers modeled 76 gigawatts of new large loads with 0.25% annual curtailment and 126 gigawatts with 1%. The study covers large flexible loads, not data centers alone, and it is a model rather than a promise about any specific grid.

The useful question is whether the project contract gives the utility enough time and control. A phased connection, storage, limited curtailment, and a clear right to delay service until the system is ready can lower the risk.

Put in writing: Publish the connection study and the project’s reliability obligations. Do not make existing customers guarantee uninterrupted growth at any cost.

Living next door

A data center is an industrial building. Treating one like an office building is how towns create avoidable fights.

What about noise, diesel exhaust, and lights?

Noise is a real problem at some badly sited or badly designed facilities. Virginia found that about 10% of operating sites in its review appeared to have generated noise complaints. Most had not, mainly because of their location or design. Ordinary noise rules often miss the low-frequency hum that bothers neighbors.

Backup generators also emit nitrogen oxides, carbon monoxide, and particles. In Northern Virginia they produced less than 4% of regional nitrogen-oxide emissions and no more than 0.1% of carbon-monoxide and particle emissions in the state review. That regional number does not answer what happens at the nearest house, which is why site-level permits and monitoring matter.

Put in writing: Require distance from homes, a third-party sound model, a low-frequency limit at the property line, testing hours for generators, downward-facing lights, a complaint process, and penalties that make the operator fix the problem.

What about farmland, transmission lines, and property values?

A campus and its power lines can permanently change a place. Virginia found that 29% of operating data-center properties in its sample sat within 200 feet of residential zoning. The same review did not find clear evidence that nearby data centers had reduced home values, but it warned that the tight Northern Virginia housing market may hide an effect.

A town does not need to wait for a property-value study to avoid obviously bad siting. It can keep industrial uses away from homes and schools, favor already-industrial or previously developed land, set height and screening rules, and require the transmission route to be shown before approval.

Put in writing: Decide where these buildings and power lines belong before a developer files. Do not create exceptions parcel by parcel next to homes, schools, farms, parks, or historic land.

What about construction traffic and emergency services?

One building usually takes 12 to 18 months to construct, and a large campus can stay under construction for five to seven years. Virginia documented heavy trucks damaging a neighborhood access road and sometimes blocking school buses and emergency vehicles. Once open, the facility has far less daily traffic, but the fire, electrical, and backup-power systems still require a specific emergency plan.

Put in writing: Set truck routes and hours, require a road-repair bond, and make the developer fund the fire, rescue, water, and inspection capacity the project adds.

Jobs and public money

The jobs are real, and so are the inflated headlines. Separate construction work, permanent work, and modeled spillover jobs.

Are the job claims inflated?

Permanent-job claims often are. Virginia found that a typical 250,000-square-foot building may have about 50 full-time workers, roughly half of them contractors. Construction is a different labor story: about 1,500 workers can be on one building at the peak, and a multi-building campus can provide years of work.

The International Brotherhood of Electrical Workers is actively organizing for these projects. IBEW says electrical subcontracting accounts for 45% to 70% of a typical data center’s construction budget, and some projects require several times a local union’s membership. The union also warns that the work will not last forever. That is the honest case: a large skilled-trades buildout, followed by a much smaller group of well-paid operating jobs.

Put in writing: Write prevailing wages, paid apprenticeships, local recruitment, and quarterly job reporting into the agreement. Publish construction jobs and permanent jobs as separate numbers.

Why give tax breaks to a building with few permanent workers?

Do not judge the deal by the investment headline. Judge the taxes the public will actually collect after exemptions, the infrastructure the public will pay for, and the risk left with residents. Virginia’s state sales-tax exemption was worth $928 million in fiscal 2023. Across five mature Virginia markets, data-center revenue ranged from less than 1% to 31% of local revenue at the time of the state review.

Some local deals have produced visible value. Loudoun attributes 38% of its fiscal 2026 General Fund to data centers. Henrico used data-center taxes to create a $60 million affordable-homeownership fund. A bad tax deal can still erase much of the upside.

Put in writing: Publish a 20-year tax schedule, every exemption, expected public costs, and clawbacks before the vote. Do not let a nondisclosure agreement hide the basic economics.

Do communities with data centers actually do better?

There are good signs, but no clean national proof of cause and effect. An August 2026 a16z article summarized Wells Fargo county data showing that counties with operating data centers had more housing, higher home values, lower unemployment, and more job growth since 2024. The article also says the comparison is skewed by places such as wealthy Loudoun County and fast-building Texas. Treat the chart as a lead, not a verdict.

The wage evidence is stronger but still needs a label. The same a16z piece used Indeed postings to show data-center pay premiums from 10% for electrical engineers to 64% for facilities managers. A Dallas Fed contact reported $45 per hour plus $150 per diem for data-center concrete work, compared with $28 to $32 elsewhere. These figures show a tight labor market; they do not guarantee that local residents get the jobs.

The best proof remains local and named: an audited budget, a water meter, a wage agreement, a road built, or a tax-funded program that residents can see.

Put in writing: Make the project report local hiring, wages, taxes, water, power, and public costs every year. Compare the promise with the result.

What local people can do

The town has the most power before the rezoning, tax agreement, water contract, and grid commitment are final.

Can residents really change the deal?

Yes. Ohio’s consumer advocate credits public engagement with helping produce the AEP settlement that shifts more unused-capacity risk to large data centers. Virginia now requires sound and utility assessments for new facilities of 100 megawatts or more and lets localities ask for water, farmland, park, historic-site, and forest analysis. Local governments still control many zoning, tax, road, and water decisions.

Timing matters. Once a town has rezoned the land, promised a tax package, and started utility work, its bargaining power falls. Residents should ask for the documents before the vote and insist that the standards are conditions of approval, not voluntary promises in a press release.

Put in writing: Ask one question at every meeting: where is this promise written, who measures it, and what happens if the company misses it?

03

Ten American examples

These towns put water, taxes, roads, housing, or utility costs into public agreements with results that can be checked.

01Water + tax base

Quincy, Washington

Microsoft + others

70%

lower city levy rate than in 2006

Quincy cut its city levy rate and built a reuse system for drinking water.

Washington’s Data Center Workgroup found Quincy’s city levy rate fell from $3.12 per $1,000 in 2006 to $0.87788 in 2025. The city and Microsoft also built a reuse system that Microsoft reports has cut its local potable-water use by 97%.

02Water-smart design

Red Oak, Texas

Compass Datacenters

0

city water used for cooling, operator-reported

Red Oak’s campus uses an air-cooled system instead of city water for cooling.

Compass says its air-cooled, closed-loop design has used no city water for cooling since the company’s founding. Red Oak’s own records document the 165-acre campus and its dedicated 252-megawatt substation.

03Recycled water

Douglas County, Georgia

Google

100%

of cooling needs served by recycled water

Douglas County sends treated wastewater to Google’s campus.

Google and the local water authority built a system that has supplied recycled industrial water to the campus since 2012. The American Society of Civil Engineers independently describes the working treatment and pipeline system.

04Local revenue

Loudoun County, Virginia

AWS + many others

38%

of the FY2026 General Fund

Data centers now provide 38% of Loudoun’s General Fund.

Loudoun’s adopted FY2026 budget attributes 38% of General Fund revenue to data centers. The county also reports that it lowered the real-property tax rate every year from 2016 through 2026.

05Infrastructure

New Albany, Ohio

Meta, Google, AWS, Microsoft + others

800+

additional acres opened for employers

New Albany used data-center revenue to open more land for other employers.

The city says 15 companies operate more than 40 facilities. Revenue from Meta supported road, water, and sewer expansion that opened over 800 acres, while preapproved formulas make school revenue predictable.

06Housing + taxes

Henrico County, Virginia

Meta + others

$60M

housing fund financed by data-center taxes

Henrico put $60 million of data-center taxes into affordable homeownership.

Henrico’s audited report lists the company holding Meta’s campus as its largest property taxpayer, with $2.36 billion in assessed value in 2025. The county used data-center taxes to create a $60 million affordable-homeownership fund.

07Tax base + connectivity

Council Bluffs, Iowa

Google

20+ mi²

planned free community Wi-Fi coverage

Council Bluffs paired a major taxpayer with free community Wi-Fi.

Three Google entities ranked among the city’s ten largest taxpayers in the 2025 financial report. Google is also a partner in BLink, a free community Wi-Fi network built without tax dollars.

08Rural investment

Maiden, North Carolina

Apple

$1B

initial investment

Apple became Catawba County’s largest taxpayer after a $1 billion investment.

Catawba County reports that Apple became its largest taxpayer after an initial $1 billion investment supporting 150 direct jobs and roughly 250 contract workers. Apple later committed another $448 million and added local solar generation.

09Water accounting

Mesa, Arizona

Meta

1.79B gal

long-term water-storage credits transferred

Mesa required water credits for 20 years of expected use.

Mesa’s financial report says Meta transferred enough long-term storage credits to offset 20 years of anticipated water use. The city also projected more than $27 million in construction and electricity sales taxes over ten years.

10Industrial renewal

Lenoir, North Carolina

Google

$1B

new investment announced in 2026

Google returned to Lenoir with another $1 billion after 17 years.

Caldwell County says Google’s Lenoir campus has supported hundreds of jobs over 17 years. In 2026 Google announced another $1 billion, a $2 million local energy-impact fund, and workforce grants; a prior expansion included $6.8 million for water-flow improvements.

04

Here are the sources

There are 25 references here. I show who published each one. Company, union, and industry-funded sources are labeled.

03A

Investment and the tax base

Most local economic value comes from construction spending and taxes. A typical facility does not employ hundreds of people forever.

PublicVirginia Joint Legislative Audit and Review Commission, 2024

Virginia’s watchdog found substantial economic benefits

JLARC estimates 74,000 jobs, $5.5 billion in labor income, and $9.1 billion in annual state GDP, while noting that most benefits come during construction.

Open source
PublicU.S. Census Bureau data processed by Our World in Data

U.S. building spend has risen sixteenfold since 2014

Inflation-adjusted monthly spending on data-center buildings exceeded $2.4 billion in January 2026. The series excludes servers, land, and maintenance.

Open source
PublicLoudoun County adopted budget

Loudoun now gets more than one-third of its general fund from data centers

The county’s FY2026 budget attributes 38% of General Fund revenue to data centers and uses a stabilization fund to manage volatility.

Open source
PublicBusiness Oregon, 2025

Oregon reports $4.6 billion in annual statewide output

The state economic-development agency estimates more than 23,000 direct, indirect, and induced jobs and about $182 million in state income taxes.

Open source
CommissionedPwC for the Data Center Coalition, 2025

The broad national estimate is large—and industry-funded

PwC estimates the sector supported 4.7 million direct, indirect, and induced jobs and $727 billion in GDP in 2023. These are economic-contribution estimates, not permanent jobs at data-center sites.

Open source

03B

Work and wages

Construction creates far more jobs than operation. Electricians, concrete workers, technicians, and other trades are in short supply, and some employers are paying more.

IndependentFederal Reserve Bank of Dallas, August 2026

Texas contractors report a roughly 50% concrete-worker premium

A Dallas Fed contact reported data centers offering $45 per hour plus $150 per diem, compared with $28–$32 per hour elsewhere in heavy industrial construction.

Open source
IndependentADP Research, August 2026

Pay gains are strongest in goods-producing industries

ADP’s July 2026 payroll data found the largest job-switcher pay gains in goods-producing sectors, where specialized skills remain scarce.

Open source
SynthesisIndeed data summarized by a16z, August 2026

Posted-pay data shows premiums across common roles

An a16z synthesis of Indeed postings found data-center premiums ranging from 10% for electrical engineers to 64% for facilities managers. Useful signal; not a controlled wage study.

Open source
UnionInternational Brotherhood of Electrical Workers, May 2025

IBEW says electrical work takes 45% to 70% of construction budgets

The union reports that some data-center projects need several times a local union’s membership. It also warns locals not to abandon their existing customers during the boom.

Open source

03C

More compute per watt

Data centers use each watt more efficiently than they did a decade ago. AI demand is still driving total electricity use sharply higher.

Peer-reviewedMasanet et al., Science, 2020

Compute grew 550% while global energy use grew 6%

A peer-reviewed reconstruction found that data-center workloads rose sharply from 2010 to 2018 while efficiency improvements held global electricity growth to roughly 6%.

Open source
PublicLawrence Berkeley National Laboratory, 2024

U.S. use was nearly flat in 2014–2016—then AI changed the curve

LBNL estimates about 60 TWh in 2014–2016, rising to 176 TWh in 2023. The same report projects 325–580 TWh by 2028 and makes the rebound explicit.

Open source
IndependentUptime Institute Global Survey, 2024

Facility overhead fell steeply, then plateaued

Industry-average power usage effectiveness fell from about 2.50 in 2007 to 1.56 in 2024. New large facilities tend to be more efficient than older legacy sites.

Open source
Commissioned451 Research for AWS, 2019

Cloud migration can cut energy and carbon for enterprise workloads

A 451 Research study estimated AWS infrastructure was 3.6 times more energy-efficient than surveyed U.S. enterprise data centers and could reduce associated carbon by 88%.

Open source

03D

Water and useful heat

Water use depends on the cooling system and the water supply. A dry-cooled project, a recycled-water project, and a drinking-water project are not the same.

PublicLawrence Berkeley National Laboratory, 2024

The federal baseline now includes direct and power-sector water

LBNL’s congressionally requested report estimates both on-site consumption and water tied to electricity generation, giving communities a better basis for project comparison.

Open source
PublicU.S. Geological Survey

The U.S. water comparison is measurable

USGS reports 322 billion gallons per day of total U.S. withdrawals in 2015. That benchmark helps distinguish a local water constraint from the industry’s national share.

Open source
PublicEU Covenant of Mayors case study

Stockholm recovers data-center heat for district heating

The city, Stockholm Exergi, and participating data centers use an open heat-recovery program to turn server waste heat into a heating input.

Open source
PublicNew York State Energy Research and Development Authority, 2026

New York is testing the same idea for U.S. cities

NYSERDA’s DATA HEAT study examines how data centers and district energy systems can be coupled as a decarbonization resource.

Open source
CompanyRamboll project case study

Odense pipes surplus server heat into the city network

Meta’s Danish campus and district-heating partners built a working heat-recovery system that supplies homes rather than venting all surplus heat.

Open source

03E

A larger, more flexible grid

A project that arrives without enough power can raise costs and delay other customers. Put the cost, construction, and curtailment terms in the utility contract.

IndependentDuke University Nicholas Institute, 2025

Tiny amounts of flexibility unlock tens of gigawatts

Duke researchers estimate 76 GW of new load could be integrated with 0.25% annual curtailment, rising to 126 GW at 1%. The estimate covers large flexible loads, not data centers alone.

Open source
PublicU.S. Department of Energy

The federal strategy treats large loads as potential grid assets

DOE points to on-site generation, storage, demand flexibility, clean-energy deployment, and grid expansion as a coordinated response to rising electricity demand.

Open source
IndependentElectric Power Research Institute

Nine live demonstrations are testing real-time flexibility

EPRI’s DCFlex program is testing load flexibility, faster interconnection, backup power, and power quality with utilities and major technology companies.

Open source
IndependentInternational Energy Agency, 2025

AI could unlock three to four times more emissions cuts than data centers emit

In the IEA’s widespread-adoption case, known AI applications could reduce 1,400 Mt of CO₂ in 2035—three to four times projected data-center emissions. This is a scenario, not a forecast.

Open source

03F

Local control

Communities have already changed tariffs and siting law. The useful time to act is before rezoning, tax, water, and utility agreements are signed.

PublicCode of Virginia § 15.2-2209.5, 2026

Virginia now requires a site assessment for facilities of 100 MW or more

The assessment must cover sound near homes and schools. A locality may also require analysis of water, farms, parks, historic sites, and forestland, and the utility must describe substations and transmission voltage.

Open source
PublicOffice of the Ohio Consumers’ Counsel, 2026

Ohio’s new tariff makes large data centers pay for unused capacity

Qualifying AEP Ohio data centers must pay for at least 85% of contracted capacity for up to 12 years even if they use less. The rule reduces the risk that other customers inherit the bill.

Open source
PublicOffice of the Ohio Consumers’ Counsel, 2026

Ohio’s consumer advocate tells towns to question secret deals

The state office advises local officials to scrutinize nondisclosure agreements and settle infrastructure, traffic, emergency-service, noise, and long-term development questions before approval.

Open source

05

How I use the sources

1

Start with public records

Government reports, audited financial statements, and peer-reviewed research carry the most weight.

2

Keep the caveat attached

A commissioned economic study or company case can still be useful. Its funding and limits must travel with the claim.

3

Check the local contract

A national benefit does not excuse a bad local deal. The permit, tariff, tax agreement, and water contract decide who gets the benefit and who carries the cost.

Publicagency, government, or audited recordPeer-reviewedacademic journal researchIndependentresearch institution or non-advocacy analysisCommissionedpaid for by an interested company or trade groupCompanyoperator-reported case or performance claimUnionlabor organization report or claimSynthesissecondary compilation; follow through to the underlying data