The AI Construction Boom Is Hiding in Offices

If you are looking at the headline data in the government’s economic report, it is easy to miss the artificial intelligence investment boom.

Take this morning’s construction spending report. The main Census release does not have an AI data center line. Instead, data center construction is lumped together with lots of other spending on business structures in the broader “office” categories. And that can create the misleading impression that office construction is booming. It jumped one percent in April and is up 9.4 percent from a year ago.

An office construction boom would be weird given the rise of remote work and the slow growth of the U.S. workforce. So what’s really going on?

If you look beneath the hood, you can see the real engine of growth is AI. Although the data doesn’t appear in the main release, the detailed data file from Census Bureau breaks office construction spending into three buckets: general, data center, and financial.

General office construction in April was 6.3 percent lower than it was a year earlier, falling to $43.8 billion. Compared with prepandemic February 2020, it is down almost 50 percent. In other words, this is behaving as you would expect.

AI data center construction spending, on the other hand, is soaring. Compared with a year ago, it is up 28.1 percent to $50.7 billion. It now accounts for 52 percent of private office construction and 2.3 percent of all construction spending.  On a longer timeline, the growth is truly explosive. Compared with February 2020, spending is up around 420 percent.

The shift has been remarkably fast. A year earlier, in April 2025, data centers accounted for 44.5 percent of private office construction. Two years earlier, in April 2024, they were just 32.9 percent. In dollar terms, data center construction has climbed from $28.3 billion in April 2024 to $39.6 billion in April 2025 and $50.7 billion in April 2026.

Beyond the Data Center

But the AI construction boom is not just data centers. We’re also seeing it in the line items covering private power construction spending. Private power construction, which includes gas and oil, rose to a seasonally adjusted annual rate of $148.7 billion in April, up 6.0 percent from a year earlier. The more revealing figure is the electric subcategory. Electric power construction climbed to $127.1 billion, up 7.3 percent from April 2025 and 9.9 percent from two years earlier.

Electric now accounts for 85.4 percent of private power construction, up from 84.4 percent a year ago and 83.5 percent two years ago. In dollar terms, the electric category increased by $8.6 billion over the past year, slightly more than the entire $8.4 billion increase in the broader power category. That means the growth in power construction is really growth in the electric infrastructure needed for a more electricity-intensive economy, including the data centers, cooling systems, substations, transformers, and backup systems behind the AI buildout.

Outside of construction, you can also make out the AI boom in factory orders and durable goods. The advance durable-goods report showed orders excluding transportation rising 1.1 percent in April, with gains in machinery, computers and related products, communications equipment, and electrical equipment, appliances, and components. Those are precisely the categories most likely to capture the physical supply chain of the AI buildout: servers, networking gear, control systems, power-management equipment, cooling equipment, turbines, generators, and other machinery needed to turn a data-center plan into an operating facility. The categories are old, but the investment flowing through them is new.

The year-to-date figures make the point more clearly. New orders for computers and related products were up 22.3 percent from the same period last year, communications equipment orders were up 30.3 percent, machinery orders were up 9.7 percent, and electrical equipment, appliances, and components were up 6.4 percent.

Even that understates the signal because the Census excludes semiconductor new orders from the computers and electronic products order figures. Traditionally, semiconductors were carved out of the factory orders report because they had very short lead times that mostly involved products being sold from existing inventories for immediate delivery. That’s not really true with respect to the AI chips. Lead times for major AI chips remain extremely long, often 36–52 weeks or more. And the so-called hyperscalers and other large customers are placing massive, multi-year forward commitments.

Semiconductors do get included in shipments, although they are broken up across a few different categories. This is one reason the shipments line for computers and electronic products is larger than the orders line, which would otherwise seem like an anomaly.

These kinds of information gaps shadow nearly every AI-era economic release. The measurements were built for a very different economy. Understanding what is actually happening to capital formation, productivity, and growth requires working around instruments that were built first for the industrial and later the office economy.