June's construction drop was a data center print, not a slowdown

Construction starts fell 20% in June, one month after a 33.5% jump in May, and the source matters more than the swing. Cushman & Wakefield pins May's surge on large healthcare, manufacturing, utility and data-center projects, which means the headline now measures how many very large sites broke ground before the month closed rather than how much building is underway across the country.

The split runs through the report's own language: construction activity "remains uneven across sectors," Cushman & Wakefield wrote, with data centers and energy-related projects responsible for the growth while "weakness in the residential and institutional sectors remains." The sub-indices show the two-step at close range: nonresidential building starts rose 17.8% in May and gave back 9.1% in June, while nonbuilding starts, the roads-and-power side of the ledger, jumped 91.9% and then fell 37.7%.

Contractor backlogs are where the split becomes measurable. Cushman & Wakefield, quoting Associated Builders and Contractors, puts commercial and institutional backlog at 8.9 months in June, infrastructure at 10.1 months, up 7.9% from a year earlier, and data center construction at 11 months against 8.5 months for projects without that exposure—a gap the report calls evidence of the sector's outsized influence on overall pipeline strength.

An eleven-month backlog is a bidding position before it is anything else: a contractor holding that much data center work prices the next speculative industrial shell differently from one who needs the job, and the difference arrives as fewer bidders and higher numbers for everyone else. Cushman & Wakefield did not publish that reading, but the backlog gap makes it one way a single sector's strength becomes every other sector's cost.

The design pipeline agrees with the contractors: the American Institute of Architects' Commercial/Industrial Architectural Billings Index, which tracks demand for design services at architecture firms and is built to forecast nonresidential construction activity up to twelve months ahead, came in at 46.7 in June—below the 50 line separating growth from contraction and the ninth sub-50 reading in ten months. Cushman & Wakefield expects the activity it reflects "is likely to remain soft into early 2027," an outlook it calls "highly sensitive to interest rate cuts and trade policy."

ENR's Construction Industry Confidence Index held at 54 in the second quarter, a reading that registers as slight optimism and not much else, while executives "continue to characterize the broader market as fragile, with strength concentrated in data centers and energy," per the report. Cushman & Wakefield expects sentiment to stay neutral-to-slightly optimistic until costs ease or rates come down, and construction employment reached 8.322 million in the second quarter, up 0.6% year over year—the labor line that decides whether anything gets cheaper.

As this publication has argued, the property recovery runs on supply, not silicon, and the billings index is the cleanest confirmation of that available: nine sub-50 readings in ten months, from an index whose only job is to see a year forward, amount to close to the strongest evidence an industrial or apartment owner can get that a competitor's project is not coming. It forecasts twelve months out and the actual buildings arrive after that, which puts the thinness past the index's own horizon.

June's starts drop reads as good news in the sectors that live off scarcity, and Hines made the same bet in August, pivoting from buying to building on the view that a global construction freeze handed it a scarcity advantage acquisitions could not match. The starts data supports the first half of that trade; the second half, whether building is actually cheap, depends on materials, on the interest rate cuts the report says the whole outlook hinges on, and on trade policy it flags as a swing factor.

For data centers the backlog gap extends the position this publication has taken on digital infrastructure capital—that it is siting power rather than buildings, and that the binding constraint is the substation date—by putting a second constraint beside it: eleven months of contractor backlog means crews are spoken for a year out, so a developer holding land and a utility agreement is not yet in the pipeline until someone is available to build it. The distinction matters for anyone underwriting the queue as though every project in it were deliverable on the same schedule.

Carry 46.7 into the fall, then: it is the number in this report that prices the next twelve months of supply for everything that is not a data center. A move back above 50 would be the first hard evidence that the design pipeline is filling for anything outside data centers and energy, and the first thing in a year to threaten the scarcity trade. Until it comes, the contractors with eleven months of backlog have no reason to look at industrial or institutional work, and the owners of that stock have no reason to expect them to.