All updates

Census Bureau · · Revised 398 series

US construction spending, September 2026: down 1.9% on housing as data centers grow

US construction spending fell 1.9% in September 2026 to $2,186.1bn (billion dollars) at an annual rate, mainly because housing spending dropped. Data-center building rose another 8.6%, to $96.5bn.

A data update from FactIQ, written from Census Bureau figures. What is FactIQ?

Key figures

US total construction spending, September 2026, seasonally adjusted annual rate
$2,186.1bndown 1.9% from August; 2.7% below a year earlier2022-10-012026-09-01Total construction spending, seasonally adjusted annual rate
US total residential construction spending, September 2026, monthly change
5.4%fell 5.4% to $861.2bn; down 8.8% from a year earlier2022-10-012026-09-01Monthly Percent Change in Total Residential Construction Spending in the U.S. (Seasonally Adjusted)
US private data-center construction spending, September 2026, annual rate
$96.5bnup 8.6% on the month; nearly double the $50.7bn of September 20252022-10-012026-09-01Private construction spending: Data center (Office), seasonally adjusted annual rate
US private nonresidential construction spending, September 2026, annual rate
$790.6bnup 1.1% on the month2022-10-012026-09-01Private construction spending: Nonresidential (Total Private Construction), seasonally adjusted annual rate

Housing caused the September decline

Construction spending is the value of building work done each month. It is a key measure of investment in homes, businesses and public works. The figures are seasonally adjusted annual rates. That means regular seasonal swings are removed and each month is scaled up to a full year.

Total residential spending fell 5.4% to $861.2bn and is down 8.8% from a year earlier. Spending on new single-family homes ($403.4bn) and new multifamily buildings ($115.9bn) was essentially flat. Most of the drop came from home improvements, which tend to swing a lot from month to month.

Business construction kept growing, led by data centers

  • Private nonresidential spending covers business buildings such as offices, factories and power plants. It rose 1.1% to $790.6bn, up from about $745bn in March–April.
  • Data-center spending rose 8.6% to $96.5bn. That is nearly double the $50.7bn of September 2025 and up from $58.1bn in March.
  • Data centers now make up about two-thirds of private office spending, which stands at $145.6bn. Office spending has grown more than 5% a month for four straight months.
  • Private power spending rose 1.2% to $171.7bn.
  • Manufacturing construction fell 0.3% on the month to $167.2bn and is about 18% lower than a year earlier.

September roughly reverses a revised August jump

August's gain was revised up to 2.0%, from 0.9% first reported, and August now stands at $2,227.4bn. August residential spending was revised to +3.6%, from +1.1% first reported. July was revised to -0.2%. September's total is about level with July ($2,184.0bn) and 2.7% below a year earlier ($2,245.8bn). Public spending was flat at $546.8bn.

Source: Census Bureau. See all CENSUS data in FactIQ.

About FactIQ and these updates

  • What FactIQ is. FactIQ is a database of official statistics from government agencies, central banks and international organisations in the US, China, India, Europe and elsewhere. AI assistants such as Claude and ChatGPT connect to it to look up the exact figures and cite the agency that published them. It also holds what companies report in their filings and say on their earnings calls.
  • How an update is made. When Census Bureau publishes new figures, FactIQ loads them and an AI model writes a short summary of what changed. Before it is published, code checks that every number in the summary matches FactIQ's record of the release, and a second review checks that nothing is overstated. Values that look like errors in the source data are left out.
  • Where the numbers come from. Each key figure links to the full series, with its source, units and history.

More about FactIQBrowse the dataUse it in Claude or ChatGPT