America Stopped Building Offices For People

On June 2, 2026, the U.S. Census Bureau quietly released its April Construction Spending report — known to economists as the "C30 series" (the government's monthly running count of how much money is being spent putting up buildings in America, broken out by category). Buried in Table 2 was a number that, on its own, looks like an accounting curiosity: data-center construction in the United States ran at a $50.7 billion annualized rate, against $43.8 billion for general office construction.[1] For the first time since the government started tracking these categories this way, the country is spending more, in real dollars, on buildings for machines than on buildings for white-collar workers. It is not a near-tie. Data centers passed offices by roughly 16 percent — and the gap is widening, with data-center construction up 27 percent over the past year.[2]

The cubicle economy stopped being the growth category

The conventional story for the last fifteen years was that America's growth was services-led — that the next dollar of private capital was funding a software-company office in Austin, not a factory in Ohio. The April number is the cleanest single-chart refutation of that story we have seen. That next dollar is now flowing into what the industry calls "hyperscale" compute infrastructure — the giant data-center campuses built by the largest cloud operators (Amazon, Microsoft, Google, Meta), each one drawing as much electricity as a small city. Picture football-field-sized buildings in northern Virginia, Phoenix, and central Texas; the high-voltage power lines being upgraded to feed them; and the long industrial supply chain underneath — electrical switchgear, transformers, liquid-cooling equipment, specialized cleanroom build-outs, concrete, copper, and steel.[3]

Two things are happening at once

The first is obvious: training and running AI models has made raw computing power the binding constraint on every frontier tech company, and the four "hyperscalers" — Amazon, Microsoft, Google, Meta — are racing to lock in capacity ahead of the next AI training cycle. Anthropic alone announced a $50 billion partnership with Fluidstack for campuses in Texas and New York in November 2025.[4] The second is structural: the office category itself is shrinking in real terms. Remote and hybrid work, shorter and cheaper leases, and a wave of conversions from mid-tier offices (what the industry calls "Class-B") into apartments have hollowed out demand. Office construction is not collapsing — but it has flatlined, while data-center construction is doubling roughly every five years.

What this means for founders

For founders, this is more than a headline chart. It is a signal about where the next decade's industrial-scale opportunities sit. Anything that touches the data-center supply chain — the software that connects data centers to the power grid, the equipment that distributes electricity safely inside the building, liquid-cooling systems that pull heat off the servers, prefabricated building components, on-site power generation, electrical sub-stations, fiber-optic cables, physical and cyber security, and the final "commissioning" step where every system is tested before the data center goes live — has a multi-decade demand tailwind that does not depend on AI models continuing to get better. Even in a world where the AI software layer becomes a commodity, the underlying compute infrastructure has already been built and someone has to run it. That is a real-economy thesis with hard concrete and hard steel underneath it.

There is a second, harder reading. The same chart says the United States, at the margin, is no longer building for the human knowledge worker. The desk-job economy stopped being the growth category. That has implications for everything from how cities project their downtown tax bases to how universities think about white-collar career pipelines. It is also a reminder that a two-track real economy — a small set of capital-heavy winners absorbing most of the new investment, and a long tail of categories that have stopped growing — is no longer a forecast. It is the data.

The takeaway

The number that will be quoted from this release is $50.7 billion. The number that matters is the crossover.

Sources

• [1] U.S. Census Bureau — Monthly Construction Spending, April 2026 release. census.gov
• [2] Data Center Knowledge — "Data Centers Become Largest Segment of US Office Construction" (June 2026). datacenterknowledge.com
• [3] ConstructConnect — "June 2026 Data Center Report" (constructconnect.com).
• [4] Anthropic — "Anthropic invests $50 billion in American AI infrastructure" (November 12, 2025). anthropic.com
• [5] Construction Dive — "Data center construction rolls into 2026" (January 2026). constructiondive.com