Dina Powell McCormick puts 500,000 unfilled jobs on the AI infrastructure build-out within the next couple of years

Half a million unfilled jobs, over the next couple of years, is the figure Dina Powell McCormick puts on what the AI infrastructure build-out will require of the labor market. The number is large enough to function as its own argument: the physical backbone of AI expansion is constrained not by capital or by demand, but by the availability of people trained to build and operate the facilities that make the compute possible.

Powell McCormick’s estimate carries a specific and falsifiable window. “The next couple of years” is not an open-ended forecast. It is a near-term deadline attached to a build-out cycle already underway. If the gap closes, it will close through some combination of accelerated training programs, redeployment of workers from adjacent trades, and possibly the automation of some facility management tasks. If it does not close, the constraint will show up directly in timelines for the hyperscale capacity that AI model development and inference depend on.

The skills dimension compounds the headcount problem in ways that raw numbers do not fully capture. Data center roles are not interchangeable with general construction labor. Electrical systems, cooling infrastructure, and network hardware each require specific credentials and working knowledge that take time to develop. The pipeline that produces those credentials was sized for a market a fraction of the scale now being demanded. Interest in the field may not be the bottleneck. The capacity to credential and place qualified workers at the pace the sector requires is a different and harder question.

There is also a timing asymmetry that Powell McCormick’s figure implies but does not spell out. Capital for data center construction moves quickly. Workforce development does not move at the same pace. Training programs and apprenticeship pipelines operate on multi-year cycles. The lag between committing to build a facility and having the workers trained to complete it on schedule is not a gap that market signals alone can close in the time available.

That asymmetry is what makes the 500,000 figure worth tracking rather than simply noting. A workforce shortfall that might be manageable friction spread over a decade becomes a hard ceiling when build-out timelines are measured in quarters. The question Powell McCormick’s number raises is not whether demand for these roles is real. The question is whether the institutions responsible for producing qualified workers are responding at anything close to the speed that construction commitments already on the books require.

Her framing sets the clock. The bet embedded in the estimate is not simply that demand will be high. It is that demand will arrive faster than the credentialing and training systems can answer it. That gap, if it materializes at the scale she describes, does not stay contained to the labor market. It flows upstream into deployment schedules for the compute that AI applications, and the companies built on top of them, are counting on being available. A reader who returns to this number in two years will be able to assess plainly whether the forecast held.