The data center jobs that stay after the ribbon cutting

The payroll that stays

Counting jobs at a data center means asking which jobs, in what unit, over what time. A building is not a campus. A construction crew works for a few years, while a permanent employee stays on the payroll, and a contractor at the front gate is on site but not on the owner’s payroll. The table keeps the units apart, so no row should be read against another as if they measured the same thing.

Meta’s two figures work out to at least 0.10 permanent jobs per construction worker (500 divided by 5,000), which is 10 per 100, because Meta says “500 or more.” Virginia’s typical building, from the two industry-reported JLARC figures, is 0.033 (50 divided by 1,500), or 3.3 per 100. We do not repeat Quartz’s ratios for Microsoft and Lancium: they rest on construction crew counts we could not trace to a primary source, and Lancium’s 57 is a 2021 jobs commitment set against a later peak crew, which are not like measures. The rows mix projected, announced and typical numbers, so each ratio compares a row with itself and the ratios are not a league table; treat them as announcements, not audited payrolls. A county contract can set its own floor per building. The tax abatement agreement Guadalupe County, Texas voted on at its April 21, 2026 meeting requires at least 30 full-time jobs per building by year 5, and Part 7 sets that agreement beside the company’s own job figures.

Brookings, in a study updated on August 10, 2026, linked approximately 1,500 U.S. data center facilities and 52 announced-but-canceled projects to county employment data from 2003 to 2024. Its design compares labor markets with built facilities to labor markets where announced facilities were canceled, so the figures below are estimates against that comparison group and not raw growth. It estimates that data-processing employment rose 56 percent over the first decade of operations in host counties and telecommunications rose 43 percent, while wages were unchanged and home prices rose 2 to 5 percent. At a typical host county, it says, the estimates imply “roughly 100-200 jobs, depending on facility type.” That is a county-level count in affected sectors, not the payroll of one campus. Brookings lists, among critics’ points, that the associated construction jobs are temporary, and its editor’s note says earlier findings on employment, wages and home prices were revised in the update.

When a developer gives a jobs number, four follow-up questions turn it into something a county can hold them to: how many are direct employees and how many contractors, how many were hired locally, how many are in the first year versus at full build-out, and how many job-years the construction phase amounts to (workers multiplied by the years each one works). Peak crew size alone says little about job-years.

What the tax breaks cost

Tax breaks are the other half of the jobs argument, and the figures come in different units, so they need to be kept apart too. Virginia’s sales and use tax exemption for data center equipment is estimated at $1.6 billion in abated taxes for fiscal 2025 in the Commonwealth’s fiscal 2025 annual financial report (Note 30, Tax Abatements, printed page 191). The same note sets what a data center must do to qualify: a new capital investment of at least $150 million and at least 50 new jobs, shared between operator and tenants, paying at least one and one-half times the locality’s average wage (10 jobs and $70 million in a distressed locality). JLARC says data centers and their tenants saved $928.6 million in sales taxes in fiscal 2023, of which the state portion was an estimated $683 million, and calls the exemption the state’s largest economic development incentive. Virginia’s first exemption, in 2008, applied to very few localities and the statewide one dates from 2010, JLARC says; a Virginia Mercury commentary says the 2008 version was projected to cost about $1.54 million. Good Jobs First puts Virginia’s 2025 loss at almost $1.9 billion in its Good Jobs First testimony to the Rhode Island Senate, citing a different Virginia report, so the figures are shown rather than reconciled. The same testimony (March 10, 2026) says Texas is projected to lose $1.3 billion in the current fiscal year and Georgia is expected to lose $2.5 billion in fiscal 2026.

Those are annual statewide totals. They cannot be divided by a single campus’s headcount. For one project, the cleanest example is in Good Jobs First’s reporting: in 2021 Columbus, Ohio approved a $54.3 million property tax abatement for a data center later revealed to be Google’s, which Good Jobs First says is equivalent to $2.7 million for each of its 20 jobs (54.3 divided by 20 is 2.715). The abatement covers more than one year and the source does not say how many, so the figure is a total per job, not an annual one. Good Jobs First also says in the testimony that the public cost per job at data centers “can reach $1 million or more,” without giving a source or period for that range.

Brookings adds its own estimates by building type: in what it calls hyperscale counties, where cloud and AI companies build to run their own workloads, incentives were about 2 percent of total construction investment, and in colocation counties, where landlords lease space to tenants, they were 62 percent of total investment. The two shares use different bases, which is why we quote them as Brookings words them, and Brookings does not define the denominator further. Disclosure is the other gap. Good Jobs First reports that Texas lost $1 billion to data centers in fiscal 2025 and reports company names but no subsidy amounts, no job creation and not even locations.

The other column: what a locality collects

JLARC measured the second column for Virginia, and it is the figure that sets the county’s side of the bargain. Data centers pay local business personal property tax on equipment and real estate tax on the buildings. For the five Virginia localities with relatively mature data center markets, JLARC reports that data center revenue ranged from under 1 percent to 31 percent of total local revenue in fiscal 2023, and that it was 31 percent in Loudoun County and 7 percent in Prince William County, the two largest markets (printed page 16). The amount depends on the local rate: taking $150 million of taxable computer equipment, JLARC estimated that counties could collect anywhere from $10.8 million to $0.4 million over five years after accounting for different tax rates and depreciation schedules.

The rows use different units, a state sales tax in dollars, a locality’s revenue as a share of its total, and an employment count, so none divides into another. What they give a county is the set of questions to ask: what its own tax rate on computer equipment is, what share of its revenue a campus would become, and whether the 50-job floor for the state exemption is a number the developer will meet and report.

If the tenant leaves

If demand for AI computing cools, or a cheaper cluster opens in another state, the servers can leave, and the substation and the water line stay where they are. The construction workers are already gone. Who pays for infrastructure a utility built for a tenant that walks depends on the utility’s contract with that tenant, including any minimum-payment term and how long it runs. Ask for that contract term before the vote. In Dominion’s Virginia territory, Part 6 lists what the State Corporation Commission’s GS-5 order sets for contract length, minimum charges, the exit fee and collateral.

The trade a county is offered

The campuses do bring real work. Fortune reported on June 10, 2026 that Meta is committing $115 million this year to free training for data center technicians and a guaranteed job on completion, in a program aimed at the skilled-trades shortage stalling its AI build-out. A construction year is a year of paychecks. But a county deciding whether to give up tax revenue deserves the numbers that describe the trade: permanent jobs by type, tax forgiven per job over the same years, the clawback, and what happens to the empty building when the tenant moves out.

Correction, October 29, 2026: Earlier versions said a 100-megawatt campus has 800 to 1,200 construction workers and 100 to 200 permanent staff, and treated Brookings’ county-level estimate of roughly 100 to 200 jobs as a campus payroll; neither could be traced to a source and both are removed. The Virginia figures of approximately 50 permanent workers and approximately 1,500 at peak construction come from industry representatives speaking to Virginia’s legislative auditors, not from an audit finding; JLARC’s own estimate is over 8,000 jobs statewide in fiscal 2023. The page also gave Virginia’s exemption as about $1.9 billion on about $33 billion of purchases (unsourced); the Commonwealth’s financial report gives $1.6 billion, and Good Jobs First’s higher figure is shown beside it. The 2008 projection is $1.54 million, and Good Jobs First names three states above $1 billion, not four. This update replaces the Meta, Microsoft and Lancium rows, which rested on Quartz alone, with announcements from the companies and the states where they exist, drops the Microsoft and Lancium ratios, and adds the financial report’s statement that Virginia has no provision for recapturing abated taxes, which is on printed page 191 of the report. Later on October 29: Microsoft’s groundbreaking post, dated June 23, 2026, was opened, and it does say the first phase is expected to create more than 600 permanent Microsoft jobs; the typical-building staffing is worded as the industry representatives’ “may have” estimate; Brookings’ incentive shares are labeled as its estimates; and the table is tied to Guadalupe County’s 30-job-per-building floor.

Editor written. AI used for spelling and grammar only.