Private Equity Found Its AI Trade, and It Looks Like an HVAC Contractor

Aug 30

3 min read

What's New

Datacenter construction grew 46% year on year in the second quarter, faster than any other construction type, and the private equity money chasing it arrived through 529 separate transactions rather than a handful of large ones. That deal count is up 25.7% from Q1 and up 56.5% from the 338 recorded a year earlier, while deal value went the other way and eased to $28.3 billion from $29.3 billion, according to PitchBook's Q2 2026 Construction and Engineering Report, published August 28 under the subtitle Datacenters take center stage. The divergence is the story. Sponsors are doing far more deals at smaller average size, which is the signature of platform building rather than platform buying. The AI infrastructure trade has moved down the chain to the trade contractors, and it is being executed one small acquisition at a time.

Why It Matters

The AI infrastructure trade has been narrated through hyperscaler capex and specialist datacenter REITs, which are the parts of the chain that private equity cannot buy cheaply. The buildout also requires mechanical, electrical and roofing capacity, and that capacity sits in thousands of owner operated regional businesses trading at multiples that support classic consolidation. For GPs, this is one of the few places where an AI linked demand signal meets a genuinely inefficient acquisition market. For technology platforms serving sponsors, it means the diligence and reporting workload is shifting toward high volume, low value transactions where per deal cost matters more than sophistication.

Big Picture Drivers

Count is growing faster than value: Construction deal count rose 40.4% quarter on quarter to 292 while value rose only 9.4% to $15.9 billion. The gap is the arithmetic of consolidation.
HVAC is the clearest expression of the theme: 76 deals worth $3.8 billion in the first half already exceed the 63 deals recorded across all of 2025.
Electrical has stopped accelerating: 38 deals in the first half against 37 for all of 2025 at a similar dollar total, which is flat rather than growing and worth watching given electrical's centrality to the datacenter thesis.
Sponsors are selling to sponsors: The largest construction exit was Sierra Platform's $220 million sale from SE Capital to Redwood Services, a secondary buyout rather than a strategic sale.
Exit value is falling as exit count rises: 56 exits against 54 in Q1, but value down to $9.5 billion from $11.1 billion. Realisations are getting smaller even as they get more frequent.
PitchBook is explicit about where it sits: Specialty construction is its most bullish segment on the grounds of fragmentation, with specialty engineering second.

By The Numbers

46% growth in datacenter construction, year on year, the fastest of any construction category tracked.
529 deals against 338 a year earlier, a 56.5% increase in transaction volume with no matching increase in capital deployed.
$28.3 billion of deal value, down from $29.3 billion in Q1, confirming that the increase is in count alone.
76 HVAC deals worth $3.8 billion, in six months, running well ahead of last year's full year pace.
$2.4 billion for HCSS and Nemetschek's Build and Construct segment, the quarter's largest transaction, with Thoma Bravo retaining a 28% stake.
$9.5 billion of exit value against $5.6 billion, in Q2 2025, still a material improvement year on year despite the sequential decline.

Key Trends to Watch

Average deal size keeps compressing: HVAC and roofing both show counts ahead of last year's pace with smaller cheques, which means multiple arbitrage depends on integration actually working.
Electrical becomes the constraint: Flat deal activity in the trade most exposed to datacenter power infrastructure suggests either scarcity of targets or pricing that has already run.
Secondary buyouts dominate the exit path: With strategics quiet, sponsor to sponsor sales set the clearing price for the whole category, which caps the return on consolidation.
Taxonomy changes complicate comparison: PitchBook has reworked its construction and engineering categories, so year on year subcategory figures need rebasing before anyone builds a thesis on them.
Labour cost becomes the underwriting variable: Roll ups in trades work only if wage inflation stays below the synergy assumption, and the datacenter boom is bidding for the same workers.

The Wrap

The interesting move here is that the AI trade has migrated down the value chain to where private equity can actually operate. Buying a datacenter is a capital markets transaction, but buying the mechanical contractor that fits it out is an operating one, and the second is where sponsors claim an edge. The risk is that everyone has now found the same door, which is why deal count is climbing while value is not. For technology providers, the requirement is unglamorous and specific: sponsors executing dozens of small trade acquisitions per platform need consolidated reporting across many entities with poor data hygiene, and that is a plumbing problem, not an analytics one.What's New