Aggregate data suggests a construction recession, but this masks a deep split. AI-related projects like data centers and power generation are booming, while other sectors like manufacturing are contracting. This creates a K-shaped market where the overall trend is misleading.
'Put-in-place' data reflects ongoing activity and cash flow. In contrast, 'starts' data, which assigns a project's full value to its first day, provides a more predictive, 'canary in the coal mine' view of where the construction industry is heading.
Unlike past booms, financial tools allow capital for AI infrastructure to be gathered almost instantly. This creates a massive divergence between committed financing and the much slower, multi-year reality of physical construction, introducing significant financial risk if the market sours.
The massive investment in premium data centers is predicated on enterprise customers paying for high-cost AI models. However, the rise of cheaper models offering nearly the same value at a fraction of the cost poses a significant threat to the revenue projections underpinning this infrastructure buildout.
The cost composition of a new data center has inverted. Historically a 50/50 split between construction and IT gear, it's now approximately one-third physical building and two-thirds expensive equipment like advanced chips and servers, changing project economics.
The next wave of data center construction will concentrate in the Rust Belt (e.g., Ohio, Wisconsin) and the Southeast (e.g., Texas, Georgia). These regions are attractive due to land availability and robust natural gas infrastructure, crucial for building independent power generation.
The scale of the AI buildout is staggering, with data center construction starts representing one out of every four dollars spent on new non-residential building projects. This makes the entire construction sector's performance highly dependent on the continued growth of data centers.
The massive scale of modern data centers, often located in rural areas, has created a unique construction challenge. Before starting the main project, contractors must first build temporary towns or 'work camps' with RVs to house thousands of specialized laborers for the duration of the build.
Despite project announcements and 'starts' peaking around 2029, the long timeline of these mega-projects means the actual peak of physical construction work ('put-in-place') will lag significantly. The industry has nearly a decade of growing activity ahead before the boom crests.
While public backlash against data centers is a major hurdle, core complaints about noise, water, and power can be addressed with existing engineering solutions like closed-loop cooling. The real challenge is ensuring developers, not the public, bear the financial cost of implementing these technologies.