The dominant narrative about AI and employment has focused almost entirely on white-collar displacement: lawyers losing work to contract analysis tools, programmers competing against sophisticated autocomplete, financial analysts replaced by faster dashboards. That framing captures something real. But it has obscured a large and counterintuitive story developing in parallel: the AI buildout is creating a labour shortage for physical tradespeople at a scale the industry was not prepared for, and cannot easily fix.
The connection runs through infrastructure. Every large language model, every deployed AI agent, every data centre running inference at scale requires physical facilities: racks, cooling systems, electrical switchgear, power transformers, and the generators and network cabling that keep everything running. Building and maintaining those facilities requires electricians, HVAC technicians, industrial automation specialists, and construction workers. These are jobs that require physical presence, that can't be done remotely, and that are not meaningfully automatable with current technology. Demand for them has grown dramatically alongside AI investment, and the supply has not kept up.
Between 2022 and 2026, demand for robotic technicians increased by 107%, for cooling and HVAC engineers by 67%, and for industrial automation technicians by 51%, according to figures cited by CNBC. The Associated Builders and Contractors trade group estimates that nearly half a million additional workers will be needed in the sector by 2027, up from 349,000 in 2026. Those are significant shortfalls in occupations where training pipelines run years long and where roughly one in four workers globally is approaching retirement age with a smaller cohort of younger workers entering behind them.
The economics reflect the shortage in the ways economics generally do. Skilled tradespeople moving into data centre roles are seeing pay increases of 25% to 30%, with some specialist positions reaching six-figure salaries in markets where similar roles previously paid considerably less. Jensen Huang argued at the World Economic Forum in January that AI was creating well-paid jobs for the people who build chip factories and data centres. The observation is accurate as far as it goes, though it somewhat glosses over the decade-long skills gap that no amount of corporate enthusiasm resolves quickly, and that affects a different population from the displaced software engineers and analysts most AI job coverage focuses on.
Huang added an additional angle to the workforce conversation in March when he proposed what amounts to a new compensation model: giving engineers a token budget on top of their base salary, effectively paying them to deploy AI agents as personal productivity amplifiers. The idea frames AI as a kind of capital that skilled workers can wield, with the company subsidising access to that capital as a recruiting tool. Whether this translates into a durable compensation structure or remains a thought experiment is unclear. But it points toward a more textured labour model than either the "AI eliminates jobs" or "AI creates jobs" narrative typically allows for: skilled workers augmented by AI agents might command premium rates precisely because they can direct AI systems as well as they direct their own hands.
The irony running through this story is significant. The sector most disrupted by AI spending is the one AI cannot yet touch. Data centres cannot be built remotely, maintained by a language model, or staffed by a chatbot. The people who run conduit, commission cooling loops, and commission the electrical switchgear are in growing demand specifically because AI has no near-term substitute for them. That is not a permanent condition: robotics will eventually reach capability levels where more of this work can be automated. But it is the condition for the foreseeable buildout, and the skills gap is already functioning as a genuine constraint on how quickly the AI infrastructure expansion can proceed. The boom that threatens one category of employment is being built, in part, by workers it cannot replace.