Construction spending on data-center "shells" intended for AI workloads in the United States surged in July, rising at an annualized rate nearly 60% compared with July 2025.
The data
U.S. Census Bureau figures released Tuesday show that construction put in place for data centers reached an annualized pace of more than $75 billion in July. The statistic measures the value of construction activity completed during the month and covers only the building costs for the large, warehouse-like shells that will later house racks of servers, processing units, memory chips and fiber-optic connections.
Why it matters
The jump in shell construction indicates that the physical build-out to support AI was gathering momentum over the summer, even as bipartisan political scrutiny of the sector has been increasing. Building these shells is often the first visible phase of expanding data‑center capacity for AI workloads.
Reality check: construction is a fraction of total cost
Despite the headline numbers, shell construction represents only a portion of a data center’s total deployment cost. Estimates put construction at roughly 20% of all-in costs; the more expensive and ongoing outlays are for processors, memory and other server hardware required to fill the facilities.
Bottom line
July’s data point signals an accelerating wave of investment in AI‑ready data-center infrastructure in the U.S., with annualized shell construction spending above $75 billion and nearly 60% higher than a year earlier, while the larger hardware spend remains the dominant component of total project costs.
