The Infrastructure File — The Tax Base and the Ballot
The Offer and the Answer: What the Ballot Box Did to the Tax Base
On November 3 the voters limited the data-center buildout in forty of forty-one local decisions. The offers those decisions answered are also in the record: the tax revenue projected, the jobs counted, and the one county that said yes for twenty-five years and now collects $1.2 billion a year. The desk files the two records side by side.
The News in the File
Ballotpedia records 41 local ballot measures on data-center policy in ten states, tallied after the election: on 28 a yes vote would limit development through bans or moratoriums, and on 13 a yes vote would facilitate it. Voters approved 27 of the limiting measures and defeated all 13 of the facilitating ones. The single exception is Harwood, North Dakota, which defeated a measure adding review requirements to industrial projects.
Ohio held twenty-one of the forty-one measures in eighteen municipalities, and the margins were not narrow. By the Cincinnati Enquirer’s accounting, eleven municipalities passed outright bans, among them Trenton, at 80 percent to 20 percent, and Sunbury, where a charter amendment banning facilities of 25 megawatts or more passed with 78.6 percent. In Perry, more than 87 percent approved financial-guarantee requirements on incentivized developers. In Wilmington, below, about 80 percent rejected each of four rezonings. Six measures decided earlier in 2026 had already run the same way: Canterbury, New Hampshire, prohibited data centers in March, 322 to 63; Monterey Park, California, in June, 10,388 to 1,372; Augusta Township, Michigan, overturned a rezoning in August, 2,314 to 269.
The ballot measures are the counted part of a larger record. Fortune counted at least 138 Ohio municipalities and townships with active moratoriums as of September 17; the National Conference of State Legislatures found lawmakers in at least 28 of the 38 states offering data-center tax incentives pursuing rollbacks this year; Data Center Watch counted some $68 billion of projects blocked or delayed in the second quarter of 2026 alone. What follows is the other half of the file: the fiscal record of the places doing the voting, the terms of the offers they answered, and the realized record of the places that answered the other way. Figures from adopted budgets and legislative audits are printed as record; figures from developers of unbuilt projects, as company figures.
The Offer, Measured
The most complete audit of what a data center offers a locality was made in Virginia, and its job count is modest. Industry representatives told the Joint Legislative Audit and Review Commission, the General Assembly’s oversight arm, reporting on December 9, 2024, that a typical 250,000-square-foot data center employs approximately 50 full-time workers, about half of them contract workers. Construction is the larger employment event: a building takes twelve to eighteen months to raise, with approximately 1,500 workers on site at the peak, and the jobs tend to be high-paying. Statewide, the commission estimated the industry contributes 74,000 jobs and $9.1 billion in gross domestic product each year, most of the benefit deriving from construction. A Brookings study of counties receiving a first large data center found private employment up 4 to 5 percent over five to six years; estimates ignoring pre-existing trends overstate the effect by about three times.
The revenue side of the audit is larger than the jobs side, and varies by two orders of magnitude between localities. In the five Virginia localities with relatively mature data-center markets, data-center revenue ranged from less than 1 percent of total local revenue to 31 percent. The commission’s sentence on the geography of the benefit is printed as written: localities in economically distressed areas of the state could benefit from data centers through increased local tax revenue, but these localities could have difficulty attracting the industry. Against the revenue stand the incentives: states forgo more than $3 billion a year to data-center tax exemptions, by Good Jobs First’s estimate, and the commission put Virginia’s return on its own sales-tax exemption at about 48 cents recovered per dollar forgone.
Wilmington, Ohio: Four Votes on 546 Acres
Wilmington is a city of 12,802 people in Clinton County, in the farm country between Cincinnati and Columbus. Census figures (2020–2024 American Community Survey) put median household income at $57,199, against $80,734 nationally, and 17.5 percent of residents in poverty; Clinton County stands higher, at $70,407 and 11.9 percent. The city’s modern history has one large event in it. Its air park was the national sorting hub of Airborne Express and then of DHL; when DHL ended its United States domestic delivery service in 2009, the hub closed, at a loss the Greater Ohio Policy Center puts at 16,000 jobs in the area. Air cargo returned in time—Amazon Air operates at the air park today—and the population stands 138 above its 2020 count.
Two data-center rezonings reached Wilmington in 2026, and only one of them reached the voters. On September 3 the council, acting by emergency ordinances effective immediately and not subject to referendum under Ohio law, rezoned the 471-acre Cosler Farm on U.S. 68 for a proposed $4 billion Amazon Web Services campus of about 1.9 million square feet. Separately, the council rezoned four parcels, about 546 acres, from rural residential to light industrial for a project of Ardent, affiliated with TAC Data Centers; the four would have taken the city’s land zoned for data centers from about 500 acres to more than 1,000. A residents’ group collected 1,529 signatures in about twenty-five days and placed the four on the ballot. On November 3, each was rejected by about 80 percent of the voters; the first ordinance drew 739 in favor and 2,885 against, in the Wilmington News Journal’s county totals, unofficial pending certification. The four parcels remain rural residential. The emergency rezoning was not decided on November 3 and is not affected by the vote.
Three Michigan Townships
Augusta Township, in Washtenaw County southeast of Ann Arbor, is a farming community of 7,345 people, and its census figures are not a distressed place’s figures. The Census Bureau puts its median household income at $108,457, above its county’s $89,180. In July 2025 the board voted unanimously, against its own planning commission’s recommendation, to rezone 522 acres of farmland for a Thor Equities data center reported at $1 billion; officials pointed at the time to millions of dollars a year in new tax revenue for the township and its schools. Opponents needed 561 signatures to force a vote and turned in nearly twice that within about a month. On August 4, 2026, the voters overturned the rezoning, 2,314 to 269. Residents have since filed petitions to recall the board members who approved it.
Delhi Township, south of Lansing, answered an offer that had not yet become an application. The township has 28,431 people, a median household income of $90,437, per the Census Bureau’s tables, and, by the Lansing State Journal’s account, annual tax revenue of $23 million; a twelve-month moratorium adopted in April runs to April 2027. On October 27 the Lansing Economic Area Partnership and Lighthouse Sustainability, a Colorado developer, announced a proposed 700-megawatt data center on 423 acres, an investment stated above $10 billion, without prior consultation with the township, by the manager’s statement. The sponsors’ projections were township tax revenue more than tripling, $23 million a year to $76 million, with $16 million a year added to the Holt and Mason schools, a $20 million community fund, and a no-cost residential solar program. On October 28 the township withdrew from the partnership. On October 30, three days after the announcement, the board voted 6 to 0 to record its opposition to any data center development in the township. No application had been filed.
Saline Township, twenty miles south of Augusta, said no and was answered in court. It is a farming community of about 2,375 people; a policy study of the case puts its annual budget at roughly $1 million. Related Digital proposed a multi-building campus there for Oracle and OpenAI on about 575 acres. On September 10, 2025, the board voted 4 to 1 to deny the rezoning; the developer sued two days later; on October 15 a consent judgment rezoned the site and bound the township to approve a twelve-year, 50 percent tax abatement whose final value awaits the State Tax Commission. The settlement’s tax-exemption form lists $1.31 billion of real property and $3.5 billion of equipment, $4.81 billion in all. The company’s 2026 statements value the project above $43 billion and put the resulting tax revenue, after the abatement, at about $147 million a year to county, schools, and township—well north of $150 million a year for the better part of two decades, in Oracle’s release—with more than 2,500 union construction jobs and about 450 permanent on-site jobs, all company figures. Construction began June 1, 2026; by late September the company reported one million union craft hours worked. The township’s share, the company states, is more than 2,000 times what the land would have yielded as farmland.
The Places That Said Yes
Pike County, Ohio, holds the poorest census figures in this file, and it signed the largest agreement in it. The county, in the Appalachian foothills, has 27,061 people, 27 below its 2020 count; the Census Bureau puts median household income at $52,736 and 19.1 percent of residents in poverty. Its largest site is the decommissioned Portsmouth Gaseous Diffusion Plant, whose cleanup has employed the county for a generation. On August 17, 2026, SB Energy, OpenAI, and Nvidia announced the PORTS-Pike Technology Campus on that ground: SB Energy will build, own, and operate the campus and lease approximately eight gigawatts of capacity to OpenAI for twenty years, with Nvidia the exclusive supplier of the computing infrastructure and $1.5 billion invested in SB Energy. The companies’ stated terms are 35,000 construction jobs across a six-year buildout through 2032 and 2,500 long-term operating jobs, capacity phased from 2028, the developer paying the full cost of the grid upgrades required, at least $4.2 billion of regional grid infrastructure built with AEP Ohio, and community funds of $40 million from each company. Every figure in this paragraph is a company projection attached to an agreement signed in August. Pike County’s permanent-jobs figure, 2,500 across eight gigawatts, stands beside the Virginia auditors’ figure for one typical facility, about fifty; the difference is the difference between a campus and a building.
Massena, New York, voted the other way on the only question put to it, by three votes to two. The town, on the St. Lawrence River, has 12,058 people; the Census Bureau puts its median household income at $54,595 and its poverty rate at 23.1 percent, the highest in this file. Its industrial ground is the former Alcoa East and Reynolds Metals site, shuttered aluminum works. North Country Colocation Services, an affiliate of the New York Digital Investment Group, proposes a 1.5-million-square-foot data center there; the company states about 200 permanent jobs, and construction estimates run from 1,800 to 2,500 workers over about two years. On September 16 the town board voted 3 to 2 against a six-month moratorium; the proposal proceeds without the pause its opponents petitioned for. Over the project stands a state question: the governor’s July 14 executive order paused state environmental permits for large data centers for up to a year; the developer states the project is exempt, and state officials have not confirmed it.
In Hermantown, Minnesota, the terms changed before the answer did. On October 30, after two meetings halted by protests and four hours of testimony running mostly against the project, the city council voted unanimously to approve a Google campus of about 1.7 million square feet and some 500 megawatts of demand. Before the vote, Google had withdrawn a request for up to $80 million in property tax abatements, reduced the project from four buildings to three, and signed a decommissioning agreement valued at $210 million. The Hermantown and Delhi sequences closed in the same week. In Delhi, the answer came three days after an announcement and before any application. In Hermantown, it came after the abatement request was withdrawn.
Loudoun County: The Realized Record
One county in the record has run the experiment to completion, and its figures are budget figures, not projections. Loudoun County, Virginia, the largest data-center market in the world, has 449,749 people, up from 312,311 in 2010; the Census Bureau puts its median household income at $181,765, the highest county figure in the nation, and its poverty rate at 4.3 percent. In fiscal 2026, by the county’s own published figures, data centers generated $1.2 billion in real and personal property tax revenue, 39 percent of the county’s overall budget. The adopted fiscal 2027 budget projects about $1.3 billion, roughly 45 percent of local tax revenue and 38 percent of the general fund by budget analyses; of the total, about $417 million is real property tax and about $879 million is personal property tax on the equipment inside the buildings, taxed at $4.15 per $100 of value. The county has cut its homeowner real property rate every year for ten years, $1.145 per $100 in 2016 to $0.805 in 2026, and attributes the cuts to the data-center revenue; its budget office states that without the industry the rate would likely exceed one dollar. In 2023 the county created a Revenue Stabilization Fund, against the stated risk that data-center revenue falls short of the budget built on it.
The county that collects the revenue has also begun, in the same years, to close the door it opened. On March 18, 2025, the Board of Supervisors voted 7 to 2 to end by-right development; new projects require a special exception. In September 2026 the board voted 7 to 1 to 1 to draft a twelve-month pause on new data-center and substation applications. The two halves of the national record meet in these figures and are printed together, without a sentence between them: Loudoun County’s median household income is $181,765, and it is drafting a pause. Wilmington’s median household income is $57,199, its poverty rate is 17.5 percent, and it voted four times, by about four to one, to keep 546 acres in rural residential zoning. Massena’s poverty rate is 23.1 percent, and its board voted 3 to 2 to let a proposal proceed. Pike County’s poverty rate is 19.1 percent, and its ground holds a signed agreement for eight gigawatts. The log proceeds to the Evening Watch.
The Cocked Hat
Three bearings never give a point; they give a small triangle, and this paper prints the triangle. The national count carries one: Ballotpedia’s verified list holds 41 measures decided on November 3, another tracker lists 47 for the same day, the inclusion rules separating the lists are not published, and the desk prints Ballotpedia’s 41 with the range noted. The DHL figure carries one: the Greater Ohio Policy Center puts the 2009 loss at 16,000 jobs in the area, while county accounts describe thousands eliminated; the desk prints the policy center’s figure as its figure. Saline Township carries the widest triangle in the file: the settlement’s filing values the project at $4.81 billion, the company’s 2026 statements above $43 billion, the $147 million annual tax figure is the company’s arithmetic, the abatement it assumes has not received its final value from the State Tax Commission, and the division of the revenue among county, schools, and township has not been published. Two categories of number run through this report and are not interchangeable: Delhi’s $76 million and Pike County’s 35,000 construction jobs are sponsor projections for unbuilt projects, the second stated across a six-year buildout without a statement of whether it counts positions or job-years; Loudoun’s $1.2 billion is an adopted budget’s accounting of money collected. The Ohio results are unofficial until the county boards certify them. Against the triangle, the single-valued facts stand: the votes were cast as counted above; the census figures are the bureau’s; the Virginia audit’s typical facility employs about fifty people; and in every locality in this file, the offer and the answer are both matters of public record.