Data centres do create jobs. The bill is the part we rarely audit
A data centre can absorb billions in capital and, once running, employ only a few dozen staff. That is not a failure of the model. It is the model. The authors of a 2026 study of 1,584 US facilities describe the same combination: heavy spending on land, buildings, computing equipment and power infrastructure, but very few people needed to run the place.
That single fact should reset the debate. Data centres are not jobless. The credible evidence says they can create local employment, but the effect is modest, delayed and uneven, with much of the gain landing in support industries rather than inside the facility. The jobs case used to sell projects is still systematically overstated. Local decisions should therefore rest on net local value, not on the largest number a press release can carry.
What the evidence actually shows
According to the authors’ account in The Conversation, they assembled facility-level information on 1,584 US data centres that opened between 2000 and 2023. They matched those sites with county data on employment, wages and business numbers, then compared host counties with similar counties that did not receive a data centre. Operators do not choose locations at random. They look for dependable power, fibre links, available land and accommodating local rules, often in places already growing. Some of the jobs might have arrived anyway.
The timing is the first surprise. In the first two years after a county’s first data centre opened, the authors found little change in employment outside construction. Three to five years after opening, host counties had roughly 1,000 to 1,300 additional nonconstruction jobs compared with similar counties without a facility. That is a real gain, but it is a county average, not a promise that a new campus will transform a local labour market.
The per-megawatt figures are more sober still. Five years after new capacity is installed, each additional megawatt was associated with about 7.4 more nonconstruction local jobs. Part of that apparent gain reflects further expansion: each megawatt installed was associated with another 1.7 megawatts of capacity over the following years. Once later growth is counted, the estimate falls to about 2.7 jobs per megawatt of eventual capacity. A 100-megawatt facility was correlated with roughly 275 additional local jobs outside the data centre itself. Those are not nothing. They are also not the thousands that often dominate announcements.
The composition explains why. The authors found employment gains concentrated in transportation and warehousing, professional and technical services, and real estate. They found little evidence of job growth in restaurants, shops and other local services. A data centre may support electricians, security firms, logistics and property services. It does not automatically create a broad service economy around the fence line.
The multiplier that outruns the evidence
The gap between that picture and the loudest claims is largely a category problem. Business Insider reported that a PwC study sponsored by the Data Center Coalition attributed 4.7 million jobs to the data centre industry in 2023, up 60 per cent from 2017. Only 603,900 of those were direct industry jobs. The rest were indirect or induced. Nathan Jensen of the University of Texas questioned the study’s 7.5 multiplier and said a multiplier of one or two was likely more accurate. The multiplier remains contested.
Gross multiplier studies are not necessarily dishonest. They estimate knock-on activity across a national or regional economy. But a national supported-employment total cannot tell a county whether its ratepayers will be better off. A 2026 review published in Smart and Sustainable Built Environment argues that several prominent input-output studies in its reviewed corpus were commissioned by parties with a direct stake in the result, and that gross multipliers are attractive in that setting because they are large, fit a headline and require no counterfactual. That is an argument about incentives, not proof that every sponsor is manipulating the result. It is reason enough to stop treating a gross number as a local forecast.
The announcement stage compounds the problem. Business Insider reported that when President Donald Trump announced the Stargate project in January 2025, he said it would create more than 100,000 American jobs almost immediately, and the publication found little evidence for the claim. In Ohio, Governor Mike DeWine said Amazon’s US$10 billion data centre expansion would support thousands of jobs. State filings showed a commitment to 1,058 new jobs, none required to be full-time, salaried positions. The contrast is not a technicality. It is the difference between a claim built for a podium and an obligation that can be counted.
Real gains, real costs
The strongest counterargument is that the sceptics understate the benefits. On this evidence, they do. NBER Working Paper 35194 reports positive estimated effects after the modern data centre expansion on total employment, construction employment, establishments, payroll, tax returns, adjusted gross income and wages. It finds persistent gains in employment and establishments, with construction effects stronger at shorter horizons. An Academy of Management Proceedings abstract reports a 3.5 per cent increase in local employment, with gains in wages, business establishments and median household income. These are not the findings of a no-jobs story.
But the same studies point to distribution and cost. The Academy abstract finds the effects concentrated in metropolitan areas, with negligible gains in non-metro counties. It also reports evidence of higher electricity prices for local ratepayers where the research design is cleaner. The NBER paper reports sustained increases in house prices and public-supply water withdrawals, and upward pressure on electricity prices that emerges later. Its authors describe the pattern as consistent with a hypothesis that data centres increase economic activity while putting pressure on house prices and electricity prices and increasing water demand. A community can gain wages and business activity and still lose on the household budget.
Subsidies sharpen the distributional question. That 2026 review also cites a Good Jobs First audit of eleven data centre megadeals that found an average of roughly US$1.95 million in subsidy per permanent job promised. The word promised matters. That figure is not a full cost-benefit analysis, and it is calculated against the promise rather than the delivered outcome. It is still a useful warning. If a community is paying seven figures per promised permanent job, it should know what it is buying, who will fill the roles and what happens if the payroll never arrives.
Consider two conditional cases, not predictions. A metropolitan county with existing power, fibre and modest incentives might see measurable employment and business activity, though housing and distributional costs would still need watching. A non-metro county with large electricity and water demands and a large subsidy relative to its tax base could face a weaker bargain: the Academy abstract found negligible non-metro gains, the cost evidence points to ratepayers, and the permanent workforce may be small. Local conditions decide the outcome. That is why a generic multiplier cannot decide the deal.
A net-value test, not a jobs slogan
The AI caveat cuts in the same direction. The Conversation authors warn that their study covers facilities opened through 2023, before the current wave of massive AI data centres, and that larger facilities do not necessarily create proportionally more jobs. That is no reason to assume AI campuses will be worse. It is reason to refuse the assumption that a bigger cheque buys a bigger local payroll. If future facilities are larger and more automated, jobs per megawatt could fall; if not, the old estimates still need project-level verification.
What should local leaders do instead of arguing over the headline? They should demand four accounts. First, an employment ledger that separates construction from permanent jobs, and direct from indirect and induced employment, with project-level headcount and wage data. Second, a fiscal ledger that places tax abatements and public infrastructure spending beside expected tax receipts. Third, an infrastructure ledger covering electricity, water and housing, stating who pays for upgrades and who bears the risk of shortage. Fourth, a delivery mechanism: public reporting against milestones, with incentives tied to verified outcomes rather than announced intentions.
Subsidy per permanent job promised is a useful starting metric. Subsidy per permanent job delivered, and per local resident employed, would be better. Neither replaces a full assessment of tax base, utility costs and environmental burden, but both discipline the conversation. A project that fails that accounting may still suit the operator. That does not make it suit the county.
The conclusion is conditional, not hostile. Data centres can be worth hosting where grid and water capacity are adequate, the tax-base gain is real, the incentive is modest and costs are allocated fairly. The evidence does not support a blanket refusal, just as it does not support a blank cheque. It supports a harder question: what is the community getting, when will it arrive, who will capture the benefit and who will pay for the infrastructure?
The jobs debate has spent too long on whether data centres create any employment. The better question is whether the public is paying too much for the jobs that actually arrive. A data centre is a machine for concentrating capital. Public oversight should concentrate on the bill.
When a county weighs a data centre against its electricity, water and tax bills, the hard part is not finding a number but deciding which number has been earned. Cloud services depend on layered infrastructure, and cloud security examines how shared systems, access and responsibility are separated and checked rather than assumed. That wider view helps a reader ask better questions about what a facility provides, what it consumes and who can show the evidence. Readers who want to understand that infrastructure can explore The Hack Academy’s online cybersecurity training programme, which lists cloud security among its subject areas, with self-paced learning and practical exercises alongside the theory.