The Great Rollover: Everything But Data Centers

Real construction spending fell across nearly every category in May. Total real spending is down 0.9 percent month over month and 5.4 percent year over year. What is worth watching is the one category that had been holding the broader numbers up, data centers have also turned negative on a monthly basis for the first time this cycle.

 

Where is the pullback concentrated?

Virtually everywhere. Nonresidential is down 1.0 percent month over month and 7.6 percent year over year. Manufacturing is the worst category, down 2.4 percent month over month and 25 percent year over year. Office (ex-data centers) is down 1.2 percent month over month and 15 percent year over year. Commercial, warehouse, lodging, and both private and public education categories are also negative. Multifamily and infrastructure are also negative year over year, down 0.8 percent and 1.9 percent respectively, but far more modestly than the categories above. They are softening, not collapsing.

What about data centers?

Year over year, data centers are still up 18.2 percent, the only category with meaningful growth on that basis. But month over month, data centers were down 0.5 percent. That's the first negative monthly print in this expansion. If that trend continues, the lone segment offsetting weakness everywhere else stops doing its job and the flat nominal headline starts showing real weakness too.

This landed the same week as the June jobs report and May JOLTS. Does the labor data connect?

JOLTS showed openings, hires, quits, and layoffs all essentially unchanged from April. The jobs report showed payroll growth in line with trend, but participation dropped and construction employment showed no real movement again. Labor availability is flat, not tight and not loose.

When you overlay the labor data onto the spending pullback, the read changes. If real volume is contracting while labor supply stays flat, the constraint isn't finding workers. It's a shortage of work. That points toward demand pulling back, distinct from the margin compression thesis built on input costs outpacing bid prices. This shifts the risk calculus. The margin compression thesis assumed volume held steady while input costs outpaced bid prices. A genuine demand pullback adds volume risk on top of that, a harder problem for contractors than pricing lag alone.

Do the leading data support this or is this a one-month read?

The Architecture Billings Index backs it up. May ABI came in at 44.5, the lowest since January, with the inquiries index dropping below 50 for the first time in four months and newly signed design contract value at its weakest level since January. ABI leads nonresidential construction spending by roughly 9 to 12 months, so this isn't a one-month signal. It points to continued softness in nonresidential volume well into 2027.

What does this mean for health care?

Private health care spending is down 0.9 percent month over month and 9.6 percent year over year. Public health care is close to flat, down 0.1 percent month over month, up 0.8 percent year over year. The private side, which is more relevant to most owners in this space, is contracting in line with the rest of nonresidential. ABI's institutional sector, which leads health care design work, sits at 46.9, the least soft of the four ABI sectors but still below the 50 threshold for growth.

What's the call here?

The pipeline isn't empty, it's narrowing. Data centers and institutional work are the two segments still showing real activity, even with data centers posting their first negative month. Everything else, manufacturing, office, commercial, lodging, warehouse, is contracting hard.

Watch next month's data center print closely. A second negative month would mean the one segment absorbing the broader pullback has stopped doing that job, and the trickle narrows further. This means shifting business development toward data center and institutional pursuits now, rather than holding out for a rebound in office or commercial work.