Lesson 6 of 9 · about 15 minutes
Jobs: Construction Crews and Permanent Staff
You'll be able to
- Tell construction jobs from permanent operating jobs and job-years from jobs
- Compute incentive cost per permanent job with the math shown
- Name the strongest evidence for and against data center job claims
- Check whether a job claim states its unit, its time period and who produced it
Data centers create a lot of jobs for a short time and a modest number for a long time. Almost every dispute about job claims comes from blurring those two.
Virginia's JLARC found in 2024 that each data center building takes about 12 to 18 months to build, with up to roughly 1,500 workers on site at peak. Once it opens, a typical 250,000-square-foot building has about 50 full-time workers, roughly half of them contractors. Those jobs tend to pay well: electricians, mechanical technicians, network engineers, security.
Statewide, JLARC estimated the industry supports about 74,000 jobs, $5.5 billion in labor income and $9.1 billion in GDP a year in Virginia. The same report said most of those benefits come from the construction phase. In Virginia, construction never stopped, because new buildings kept starting. A county with one campus will see a construction wave and then a much smaller permanent staff.
Units that get mixed up
| Phrase you hear | What it might really mean | Ask |
|---|---|---|
| "2,000 jobs" | Peak construction headcount for a few months | Peak or average? How many months? |
| "3,000 jobs" | Job-years over a multi-year build | Jobs or job-years? |
| "500 jobs supported" | Direct plus indirect plus induced, from a model | Which model? Who paid for the study? |
| "50 permanent jobs" | On-site staff, may include contractors | Employees or contractors? Local hires? |
Example: turning a headcount into job-years
A campus has three buildings, each built over 16 months. Average crew is 900 (peak 1,500). Buildings overlap partly, so total construction runs 30 months.
- One building: 900 workers x 16 months = 14,400 worker-months = 1,200 job-years.
- Three buildings: 3,600 job-years.
- Spread over 30 months (2.5 years): 3,600 / 2.5 = an average of 1,440 people working at once.
So "3,600 jobs" (job-years), "1,500 workers at peak" and "1,440 on average" are all accurate descriptions of the same work. Whichever one you repeat, label it.
Cost per job
Cost per job is the incentive value divided by the jobs. It is simple and it is blunt, so say exactly what went into it.
Example: one job count, priced several ways
Invented Harlow County gives up $8.8 million a year in property tax for 10 years. The state sales tax exemption is worth an estimated $30 million over the same period. The campus promises 60 permanent jobs.
- County only, per year: $8,800,000 / 60 = about $146,700 per job per year.
- County, 10 years: $88,000,000 / 60 = about $1.47 million per job.
- County plus state, 10 years: $118,000,000 / 60 = about $1.97 million per job.
For comparison, CNBC reported one Microsoft site in Illinois got more than $38 million in exemptions and created 20 permanent jobs, about $1.9 million per job when you divide one figure by the other.
Supporters object, with reason, that cost per job ignores the tax base. A data center may pay more in remaining property tax each year than the incentive costs, and its workers need few public services. The fair approach is to put cost per job next to net revenue whenever either one comes up.
Models, multipliers and "jobs supported"
Big job numbers in developer releases usually come from an input-output economic model. The model takes construction spending and operating spending, then estimates three layers: direct jobs on site, indirect jobs at suppliers, and induced jobs created when all those workers spend their pay at grocery stores and car dealers. The ratio of total jobs to direct jobs is the multiplier.
Multipliers are real economics, but they carry assumptions that the public rarely sees. The biggest one is where the money is spent. If the steel, switchgear and servers come from out of state, and half the crew lives in out-of-state hotels for 16 months, much of the spending leaks out of the county. A model built for the whole state will show more local benefit than a single rural county actually gets.
Example: how one assumption moves the answer
A study for invented Harlow County assumes $900 million of construction spending and a statewide multiplier of 1.8 on 1,200 average direct construction jobs.
- Study's total: 1,200 x 1.8 = 2,160 jobs "supported" during construction.
- Suppose only 40 percent of direct workers live in the county. Local direct jobs: 1,200 x 0.40 = 480.
- Apply a smaller county-level multiplier of 1.3 to those local jobs: 480 x 1.3 = 624.
Both numbers can be defended. One answers "what happens to the state economy," the other "what happens in this county." When the study comes up at a hearing, ask which question it answered, and who paid for it.
| Question for any job study | Why it matters |
|---|---|
| Which model and what geography (county, region, state)? | Larger areas capture more spending and show bigger multipliers |
| Who commissioned it? | Industry-funded studies are not wrong by default, but the public deserves to know |
| Jobs or job-years? | The same work can be counted either way |
| Local hiring assumption? | Traveling crews leak spending out of the area |
| Does it subtract anything? | Few studies count farm jobs or other projects displaced |
Virginia's JLARC used its own modeling for the 84-jobs-per-million figure, which makes it more independent than a developer study. In an October 2026 Cardinal News opinion piece, independent researcher Tommy Turner points to a PwC study commissioned by the Data Center Coalition, the industry trade group, as another source that circulates widely. Name the sponsor every time.
The strongest case on each side
For the job claims
JLARC's incentive reviews, as reported in a June 2026 Cardinal News opinion column by Dwayne Yancey, found the data center exemption produced 84 jobs per $1 million of exemption. JLARC's 2025 report put the average across all the incentives it reviewed at about 58. The permanent jobs pay well and do not require a four-year degree in many cases. Construction trades see sustained demand in active markets: Turner, using federal union filings, found inside electrical construction union membership grew 17.3 percent faster in hyperscale metros than in comparable ones through 2025. He built that series himself, since no published one exists, so treat it as one researcher's estimate.
Against the job claims
Permanent jobs are few for the capital involved. Idaho's exemption needs only 30 jobs for $250 million. Turner also noted that JLARC's 50-jobs figure came from data center representatives and matches Virginia's statutory threshold, so it may reflect the requirement rather than independent counts. Construction crews are often traveling specialists, so local hiring may be thinner than the announced number. And construction jobs end, which matters most in a county that will only ever host one campus.
Moses Lake, Washington, is a good check on hydro-powered industrial promises. Group14 built a 1-million-square-foot battery materials campus there and Sila a plant of more than 600,000 square feet, both citing cheap hydropower. REC Silicon halted polysilicon production there in December 2024, and OneD closed its pilot plant in May 2025, less than a year after opening it, the Moses Lake news site Source One reported. Pierson's What Power Really Costs Idaho was written before some of those closings; date your examples.
Common mix-ups
- Claim: A data center will employ thousands. What the evidence shows: During construction, yes, for a year or two per building. Operating staff is typically dozens per building.
- Claim: Data centers create no jobs. What the evidence shows: They create well-paid technical jobs and large construction employment. The count is small relative to investment.
Reporting angle
- Ask for permanent jobs, contractors and construction jobs as three separate numbers, with wages.
- Ask what share of construction workers are expected to be local, and how that will be tracked.
- Request state reports on whether incentive recipients met job targets.
- Find out who paid for any economic impact study and which model it used.
Research angle
- JLARC Data Centers in Virginia: Virginia's state audit of the industry.
- JLARC Report 604: Evaluation of economic development incentives.
- Cardinal News, data center jobs: Turner on jobs and the rest of the ledger.
- Cardinal News on the Virginia tax break: What the state study found on return per exemption dollar.
- Idaho Code 63-3622VV: The data center sales tax exemption.
- GMU CARE STEP Brief 05: Medsker and Virmani on data centers.
- CNBC on state tax forfeits: Survey of revenue states give up.
Check yourself
1. JLARC's 2024 report says a typical 250,000-square-foot data center employs about how many full-time workers?
About 50, roughly half of them contractors. Up to about 1,500 can work on site at the peak of construction.
2. A developer says a project will create '3,000 jobs.' Construction lasts 18 months with 2,000 workers at peak. What should you ask first?
Construction headcounts, job-years and permanent positions are different units. Mixing them inflates claims.
3. A county gives up $8.8 million a year in property tax for a campus with 60 permanent jobs. What is the annual cost per permanent job?
$8,800,000 / 60 = $146,667 per job per year.
4. Researcher Tommy Turner pointed out that the 50-jobs figure JLARC cited may reflect what?
He noted that 50 jobs is the exemption's threshold and that JLARC's figure came from industry representatives, so it may reflect the requirement rather than measured employment.
5. Which of these is the strongest pro-jobs evidence in the Virginia studies?
JLARC estimated 84 jobs per $1 million, versus about 58 averaged across all the incentives it reviewed in 2025, with most activity coming from construction.
Terms in this lesson
- Job-year
- One person employed full-time for one year. Two years of work by 500 people is 1,000 job-years.
- Permanent job
- An ongoing operating position, such as a technician or security officer, that lasts after construction ends.
- Indirect and induced jobs
- Jobs at suppliers (indirect) and jobs supported by workers spending their pay (induced), estimated by economic models.
- Cost per job
- Incentive value divided by jobs created. Say whether it is per year or over the life of the incentive.
Sources
- JLARC, Data Centers in Virginia (2024) checked 2026-10
- Cardinal News, Researcher: The data center jobs are real (Turner, Oct. 5, 2026) checked 2026-10
- Cardinal News, Data center tax break was Virginia's 2nd best incentive (June 19, 2026) checked 2026-10
- CNBC via NBC Los Angeles, States forfeit hundreds of millions to data centers (June 20, 2025) checked 2026-10
- JLARC, Economic Development Incentives Evaluation (2025, Report 604) checked 2026-10
- Idaho Code 63-3622VV checked 2026-10
- George Mason CARE, STEP Brief 05: Data Centers (Sept. 2026) checked 2026-10