As data centres become denser and more technically complex, Maksim Gavriliuk, Independent Digital Infrastructure and Data Centre Expert, explains why workforce planning needs to start long before handover.

A data centre can be designed in months and built in a few years. Developing an experienced operations engineer takes much longer. That difference is becoming harder to ignore as AI infrastructure expands. Most capacity discussions still revolve around megawatts, rack density and access to compute. Staffing tends to come later, even though experienced technical people are much harder to add quickly.

The shortage is measurable, and it is getting worse

Uptime Institute’s 2026 Global Data Center Survey collected responses from more than 800 data centre owners and operators. It found that more than half were having difficulty finding qualified candidates for open positions, and Uptime has warned that recruitment and retention pressures are continuing rather than easing.

Retention is part of the same problem. Uptime reports that staff turnover remains a persistent challenge, with employees often lured away by competing data centre companies. As new AI facilities come online, operators are increasingly competing for the same experienced people, as well as for power, equipment and suitable sites.

Technical roles have consistently been among the hardest to fill. Uptime’s workforce research has repeatedly identified operations, mechanical and electrical roles as particularly difficult to staff, while also highlighting growing demand for controls and monitoring specialists. Those are precisely the people who sit closest to the day-to-day operation of increasingly dense infrastructure. General technical experience helps, but it does not remove the need for people who understand how these environments behave in practice.

There is also a demographic issue. Uptime has warned for several years that mature data centre markets in the US and Western Europe face significant retirements among experienced technical staff, with particular pressure on senior operational roles.

The loss is difficult to measure simply in headcount. Experienced operators carry knowledge that rarely appears completely in procedures: how equipment normally sounds, which alarms are genuinely unusual, when a reading deserves escalation or how a system behaves after years of operation. Some of this can be documented. Some of it only develops through experience.

Uptime previously estimated that global data centre staffing requirements would increase from around two million full-time equivalents in 2019 to nearly 2.3 million by 2025. High-density and AI-related developments are now adding to demand in a market that was already struggling to recruit.

So the industry is trying to expand its workforce while some of its most experienced people are approaching retirement. Replacing the numbers alone will not replace their experience.

AI facilities need different skills

The problem is not solved simply by recruiting more people into existing roles because the work itself is changing.

Uptime’s 2026 survey found that modal rack densities are continuing to rise, with a growing number of operators reporting peak racks of 30 kW or above. AI training deployments can run well beyond that, and rack-scale architectures at the top end of the market go considerably higher again. In that part of the market, liquid cooling is becoming a core infrastructure consideration rather than a specialist exception.

That changes the operational workload. Teams may need to deal with coolant quality, leak response, cooling distribution units and equipment that connects the IT environment much more directly to the facility cooling system.

Electrical operations are changing as well. AI clusters introduce concentrated loads and operating conditions that are different from those found in many traditional enterprise data halls. At the same time, the distinction between IT and facilities responsibilities becomes less clear when a cooling issue on the facility side can immediately affect high-value compute equipment.

This is one area where I think training often lags behind design. Organisations can install new technology faster than their operating model adapts to it. A team can therefore be fully staffed on paper and still lack some of the skills required by the infrastructure it has been asked to run.

Training has to start before handover

Recruitment alone will not close this gap. Operators also need a better way to develop people already inside the industry and to bring in technicians from adjacent sectors.

Structured training is part of that. So are visible career paths. A junior technician is more likely to stay if there is a credible route into senior operations, engineering or management, rather than an expectation that shift work remains essentially the same for years.

Operational documentation matters for the same reason. Clear procedures, defined competencies and structured assessment reduce the amount of knowledge that depends on informal handover between experienced staff and new recruits.

Reviewing a data centre operations standard this year has made the same point clear to me. A significant part of the skills problem is also a knowledge-transfer problem. Organisations may have experienced people, but that experience has limited value to the next generation if it remains largely in individuals’ heads.

I have seen a similar issue in markets where data centre construction has moved faster than the development of local operations teams. The facility can be completed on schedule while the people expected to run it are still building confidence with unfamiliar systems. Closing that gap takes structured knowledge transfer, clear responsibility and time. Hiring more people at the end of the project does not solve it by itself.

Staffing belongs in capacity planning

Data centre developers already plan years ahead for power, land and long-lead equipment. Staffing should enter the planning process much earlier than it often does today.

The question at the design stage should not only be whether enough power and cooling will be available when the facility opens. Operators should also know what skills the facility will require, where those people will come from and how long they will need to become competent in the systems they will operate.

That becomes especially important with AI infrastructure because technical change is occurring faster than the industry can rely on experienced hires alone.

A data centre may be physically ready before its operating team is. At that point, the missing capacity is no longer electrical or mechanical. It is operational.

Planning for people earlier will not remove the skills shortage, but it reduces the risk of discovering it only after the infrastructure has already been built.