UK datacentre job claims are measuring different workforces

Verdant's 9 September briefing estimates roughly 4,400 direct jobs at existing UK datacentres and about 10,400 direct jobs across developments currently planned. It contrasts the latter total with a 2024 techUK report projecting 40,200 additional direct operational roles by 2035. The gap is important, but “10,400 versus 40,200” is not a like-for-like forecast contest.

The techUK figure is conditional. Its report models what could happen if annual UK datacentre-capacity growth accelerated from about 10% to 15%. Alongside 40,200 additional operational roles by 2035, it projects 18,200 additional direct construction roles over 2025–35. The report also estimates indirect and induced economic effects, gross value added and tax. Verdant instead asks how many permanent direct jobs may operate the set of projects now in the pipeline. One number is a total for a project universe; the other is an addition under a national growth scenario.

The sources also build their estimates differently. Verdant says it assembled a database from planning applications, ministerial statements and industry releases and then recalculated employment per megawatt. The Guardian's report says the analysis used staffing information for 20 projects and international comparisons. It reports Verdant's estimate of 8.6 direct jobs per megawatt for existing sites, against 43.7 attributed to the industry projection, and a possible 1.6 permanent roles per megawatt for much larger future projects. These are model inputs and extrapolations, not a census of future workers.

Four denominators hide inside two headlines

First, stock and change are different. Verdant's 10,400 is presented as a total across currently planned facilities; techUK's 40,200 is an additional number relative to its baseline. Second, job types differ. Permanent site operations should not be added casually to temporary construction work, supplier employment or jobs induced elsewhere in the economy. Third, the horizon differs: a named planning pipeline and a capacity-growth scenario to 2035 can contain different facilities. Fourth, geography differs. A UK-wide multiplier cannot show how many jobs will be accessible to residents of a particular host authority.

There is no official estimate that resolves the disagreement. In a 15 July parliamentary answer, the Department for Business and Trade said it had not made a specific estimate of permanent jobs arising from potential investment; it cited techUK's 40,200 operational and 18,200 construction figures. That makes the industry model a referenced external estimate, not a government-produced employment forecast.

The strongest counterarguments should stay visible. The government told the Guardian that jobs per megawatt confuses electricity capacity with actual consumption and omits the wider economic, security and sovereign-compute case. techUK defended its published methodology as using industry data and standard economic-impact techniques. Those points do not validate either headline, but they show why a single employment-per-power ratio cannot settle the infrastructure decision.

Both source packages disclose material limits. Verdant's landing page does not publish a machine-readable project database, and a sample of projects with public staffing data may not represent future facilities. The techUK methodology uses input-output modelling, jobs-per-megawatt estimates and economic multipliers; its annex notes constant-returns assumptions, changing market dynamics and a skew towards evidence on very large sites. Automation and economies of scale could reduce on-site labour intensity, while new services, suppliers or construction programmes could create work outside the permanent site workforce. None of the 2035 figures is an observed outcome.

Build an occupation-and-time ledger

Before funding training, workforce leaders should require one project-level ledger with separate rows for construction job-years, permanent site roles, supplier roles and induced effects. Every row should identify occupation, skill level, location, start and duration, baseline, scenario, confidence range and evidence source. Local and national totals should not be interchangeable.

That ledger changes provision decisions. Construction may create an earlier peak for electrical trades and commissioning specialists. A smaller permanent workforce may still require scarce capabilities in power systems, cooling, networking, cybersecurity and incident response. Wider digital-service growth may create roles away from the datacentre site and therefore needs a different regional training strategy.

Planning and procurement can improve the evidence by requiring developers to report realised jobs against the categories used in their applications, including subcontracting and commuting assumptions. Annual post-construction reporting would let training providers compare promised demand with vacancies and retention rather than wait for another national model.

Datacentres may still have value for resilience, research or sovereign compute. Those cases need their own evidence and distributional analysis instead of borrowing certainty from a disputed jobs total. The practical question is which work appears, where, when and for how long. The Skills Atlas can structure the capability side; sponsors must supply auditable workforce quantities.