Hundreds of billions in AI infrastructure spend, and the constraint slowing it all down is a shortage of the engineers and tradespeople who can actually build it.
A 60 MW data centre delay can cost over $14 million a month in lost revenue[1]. Right now, across the US, that kind of delay usually isn't about money. Hyperscalers have committed hundreds of billions of dollars to AI infrastructure this year alone. It's about a shortage of electricians.
Electrical work accounts for 45% to 70% of total data centre construction cost, according to the IBEW[2]. That's not an exaggeration, it's the physics of the build: switchgear, UPS systems, PDUs, backup generation. Every other trade on site sequences around the electrical scope. When that falls behind, the whole project falls behind with it.
And it's falling behind because the people aren't there, according to Deliotte's 2026 Engineering and Construction Industry Outlook[3].
Construction forecasts put the 2026 shortfall at close to half a million workers, with 41% of the existing workforce expected to retire by 2031.
Trade groups estimate the industry needs around 300,000 additional electricians just to meet current demand, at a moment when construction unemployment is sitting at a historic low, so there's no reserve pool to absorb a sudden surge in projects[4]. Operators feel it directly: nearly half report difficulty finding qualified candidates for open roles, a gap that's barely moved in years despite the industry throwing more training and recruitment strategy at the problem.
The hyperscalers themselves are the clearest evidence of how serious this is. Meta has launched a $115 million free skilled-trades academy with Texas as one of four 2026 pilot states. Microsoft has run its own Datacentre Academy with community colleges since 2019. Google has committed funding to train 100,000 electrical workers and 30,000 new apprentices through a partnership with the electrical training ALLIANCE. Amazon Web Services (AWS) runs its own apprenticeship and pre-apprenticeship pipelines straight into mission-critical roles. None of these companies are short on cash. They're building training infrastructure because they can't buy their way around a labour shortage. They can only build their way through one.
That's the part of the story that doesn't get the same coverage as the gigawatt announcements. Land, power, and tax incentives used to be the variables that decided where a data centre got built. Labour capacity now sits alongside them as a genuine constraint on where and how fast operators can build. Texas is the clearest example: it's drawing a wave of new investment and, according to JLL, is on track to overtake Northern Virginia as the world's largest data centre market by 2030[5]. That growth is colliding directly with the same labour shortage as everywhere else, which is exactly why Texas is also where Meta chose to pilot its training programme.
A licensed electrician with genuine mission-critical experience isn't interchangeable with one who's only worked commercial fit-outs, and a commissioning engineer who understands phased energisation can't be produced by raising a budget line. The wage data backs this up. Data centre construction work is commanding roughly 30% more than standard builds, and in the hottest markets, electricians in their twenties are earning well into six figures[6].
It also means the competition for this talent isn't contained to data centres. The same certified electricians, controls engineers, and mechanical specialists are being pulled toward semiconductor fabs and energy infrastructure projects running on similarly tight timelines. Every new site, wherever it's built, draws from the same shrinking, increasingly mobile pool of specialists, and the operators who move fastest aren't necessarily the ones with the biggest budgets.
This changes what "hiring well" actually means on a mission-critical build. It's not a job board problem, and it isn't solved by posting a role and waiting. The teams staying on schedule right now are the ones who built relationships with specialist recruiters before ground broke, kept a pre-qualified pipeline warm through the design phase, and treated sourcing for critical facilities engineers, commissioning specialists, and controls talent as part of the construction plan rather than something to scramble for once the timeline started slipping.
That's the gap Thor sits in. We're not generalist recruiters trying to learn mission-critical infrastructure on the job. We know the difference between a critical facilities engineer who's commissioned a hyperscale campus and one who's only run a single substation, and we have the networks built specifically around people who've done the former. On builds where a single delayed trade can cost millions a month, that distinction is the whole job.
The capital behind this boom isn't going anywhere. Whether it gets deployed on schedule depends entirely on who's available to build with it.
