AI Construction Boom Hidden in Office Data
When reviewing government economic reports, it’s easy to overlook important details, especially regarding the growing investment in artificial intelligence.
Take, for instance, the construction expenditure report released this morning. You won’t find a specific line for AI data centers. Instead, this type of construction is grouped with various other business structure expenditures under the “office” category. This can create a somewhat misleading perception that office construction is thriving. It saw a slight rise of 1% in April and an annual increase of 9.4%.
Yet, an office construction boom seems a bit odd, considering the rise of remote work and the tepid growth of the US labor force. So what’s really happening?
By digging deeper, it becomes clearer that AI is a significant growth driver. While the main release might not reflect this, the Census Bureau’s detailed files categorize office construction spending into three segments: general, data center, and financial.
General office construction in April totaled $43.8 billion, which is actually a 6.3% decline from the same month last year. In comparison to pre-pandemic February 2020, that’s nearly a 50% drop. I mean, it’s pretty much aligning with expectations.
On the flip side, expenses for AI data centers are soaring. Compared to last year, there’s been a 28.1% increase, bringing the total to $50.7 billion. Currently, this accounts for 52% of private office construction and about 2.3% of total construction spending. If you zoom out to a longer timeline, the growth is astonishing—a roughly 420% increase since February 2020.
The change is quite swift. A year back in April 2025, data centers made up 44.5% of private office construction. In April 2024, it was just 32.9%. In dollar figures, construction for data centers surged from $28.3 billion in April 2024 to $39.6 billion in April 2025, hitting $50.7 billion in April 2026.
Beyond Just Data Centers
The AI construction surge isn’t confined to data centers alone. You can also spot it in private power generation construction spending. This segment, which includes gas and oil, rose to a seasonally adjusted annual rate of $148.7 billion in April, marking a 6% increase from the previous year. The electrical subcategory notably rose to $127.1 billion—up 7.3% from April 2025 and 9.9% from two years prior.
Electricity now makes up 85.4% of private power construction, up from 84.4% a year ago and 83.5% two years ago. In monetary terms, the power sector increased by $8.6 billion over the past year, slightly outperforming the broader power sector’s $8.4 billion rise. This suggests that growth in power construction is indicative of the necessary infrastructure for a more power-heavy economy, including backup systems for data centers, cooling systems, substations, transformers, and related AI construction.
The impact of the AI boom extends beyond construction. For instance, factory orders for durable goods show a similar trend. According to recent reports, orders, excluding the transportation sector, climbed by 1.1% in April. This includes growth in machinery, computers, communications equipment, and electrical appliances—categories critical for an AI-enabled supply chain. We’re talking about servers, network devices, control systems, and other essential machinery for data centers.
When looking at year-to-date stats, the picture becomes even clearer. New orders for computers and related products have jumped by 30.3%, while machinery orders grew by 9.7%, and electrical equipment, home appliances, and parts saw a 6.4% uptick compared to last year.
Even these numbers might undervalue the situation since current statistics exclude new semiconductor orders, traditionally left out due to their short lead times. But with AI chips, those lead times often extend 36 to 52 weeks or more. This has led large customers to commit to extensive, multi-year orders.
Though semiconductors are counted in shipment figures, they get categorized in various ways. This could explain why the shipping figures for computers and electronic products sometimes eclipse the ordering figures, which might seem unusual otherwise.
This kind of discrepancy highlights a complexity in almost every economic report during the AI era. The metrics were built for very different times. To truly grasp what’s occurring regarding capital formation, productivity, and economic growth, we need to examine the equipment created initially for industry and subsequently for office settings.
