Applied Digital CEO Wes Cummins told CNBC that historically only about 10% of large-scale industrial construction projects deliver on time, and he expects the AI infrastructure space to see "pretty significant delays through 2026 and 2027," according to 24/7 Wall St. The bottleneck: labour-force constraints that Cummins says "will only intensify from here."
The warning lands at a precarious moment. Hyperscaler annual capex references jumped from approximately $400bn to nearly $600bn in just three months, with Cummins describing the magnitude of the jump as the defining feature of the cycle. If supply slips while demand surges, operators with capacity already online gain outsized pricing and renewal leverage.
Applied Digital ($APLD) itself has been an outlier, delivering facilities faster by using modular, offsite-assembly approaches and building in lower-competition labour markets like North Dakota and Louisiana rather than West Texas. The company posted Q3 revenue of $126.6m — 61% above consensus — with shares up sharply year to date.
Analysis
Cummins is saying what construction economists know but AI hype cycles ignore: building physical things is hard. The $400bn-to-$600bn capex jump in 90 days is demand sprinting ahead of the one resource you cannot spin up in software — skilled electricians and HVAC crews. Applied Digital's modular model is smart but does not scale infinitely. The winners will not be the companies that sign the biggest leases. They will be the ones that actually deliver megawatts on time.
