IT Pro published an analysis on the structural problems delaying data center construction. Between a third and half of all U.S. data centers planned for 2026 will be delayed or canceled. The factors converge: skilled labor shortages of approximately 439,000 construction workers nationwide, hyperscale sites needing 4,000 to 5,000 workers per build, grid interconnect delays now stretching 4 to 5 years, and supply chain disruptions across switchgear, transformers, and cooling components. Memory and storage costs have risen 3 to 5 times since Q1 2025 as AI buildouts crowded consumer categories out of the supply queue.
The IT Pro framing is comprehensive. The piece the cooling industry should focus on is the supply chain section, because cooling equipment lead times are now functioning as a constraint on the timing of facility commissioning even when grid power and construction labor are available.
Four categories of cooling equipment are now in extended lead time backlogs that materially affect project schedules. Chilled water plants, including chillers, pumps, and associated piping, have lead times that have stretched to 18 months at major manufacturers. Cooling distribution units for direct-to-chip applications, where demand grew sharply when AI rack densities forced the architecture shift, are quoted at 14 to 16 months for new orders at most vendors. Specialty valves and pumps for liquid cooling, especially the larger sizes needed for hyperscale, have hit 12 to 18 months. And cooling tower mechanical components, particularly fans and motors at the larger sizes, are in similar territory.
The component that gets the most attention is switchgear because it sits on the electrical critical path. The cooling components don't get as much attention because they're not the first thing operators worry about. They become the first thing operators worry about once the cooling lead times bind the schedule even after switchgear and transformers arrive.
The 439,000-worker construction shortage is the supply-side number. The harder number to find is the shortage of qualified mechanical contractors, commissioning engineers, and operators trained on the specific liquid cooling architectures that AI workloads now require. Most facility commissioning teams were trained on chilled water plants and CRAH-based air cooling. The shift to direct-to-chip and immersion architectures requires retraining, and the training pipelines are not scaled to the demand.
This shows up in project timelines as commissioning delays. A facility can have all the equipment on site, all the construction trades present, and still take an extra three to six months to commission because the small number of qualified liquid cooling commissioning engineers are booked through 2027. The lifecycle expertise gap that Salute has been calling out is the same gap that shows up in commissioning timelines for facilities that look complete on paper.
The combined picture is that any new data center project starting today should assume cooling equipment will not arrive faster than 14 months, that commissioning will require booking liquid cooling specialist engineers at least 12 months in advance, and that the construction labor pool can support the project only if other concurrent work in the region is not competing for the same trades. The 18-month timeline from groundbreaking to operational that hyperscalers were achieving in 2022 is now closer to 32 to 36 months for facilities of comparable scale.
Operators planning capacity additions for 2027 need to be placing cooling equipment orders this quarter. Operators planning for 2028 need to be locking down liquid cooling commissioning engineers now. The slip risk on AI capacity is structural, not cyclical, and the cooling supply chain is one of the two or three load-bearing constraints determining when capacity actually comes online. The hyperscalers that have already integrated cooling vendor commitments into their multi-year capex planning are going to outpace the ones that are still treating cooling as a late-stage procurement decision.