U.S. construction spending fell.
The housing detail matters.
September construction spending fell 1.9%, but new-home spending was nearly flat. A closer look separates remodeling, offices and public projects.
The November 2 release from the U.S. Census Bureau describes a smaller national construction market than a year earlier, but it does not describe every part of that market moving in the same direction. September spending ran at a seasonally adjusted annual rate of $2.1861 trillion, down from a revised $2.2274 trillion in August. Private residential work fell sharply. Private nonresidential work rose. Public construction was approximately unchanged once the survey's uncertainty is taken seriously. Those three observations belong together: using the headline alone to judge a homebuilder, a hospital contractor or a municipal road program would erase the differences that make the report useful.
The most revealing detail sits below the residential total. New single-family construction was essentially flat in September, and new multifamily construction edged up. The much larger drop in the broader residential measure therefore should not be described as an equivalent collapse in work on new houses or apartments. Improvements to existing owner-occupied homes are part of the residential universe, and the gap between the total and the two new-building categories widened and narrowed very differently. The report is a measure of construction work and related costs, not a count of houses, permits or transactions.
This analysis uses the Census release, its published tables and classification definitions. The project photographs are identified U.S. government archive images illustrating different kinds and stages of construction. They are not a Census sample list, a claim about which individual projects caused September's change, or evidence that the photographed work occurred during that month. National estimates cannot identify the contribution of an individual site from a photograph.
A $41 billion change in an annualized rate is not a $41 billion monthly cash loss
The headline compares two seasonally adjusted annual rates. The detailed table puts September at $2,186,145 million and August at $2,227,374 million, a difference of $41,229 million in annual-rate terms. Annualization places a monthly pace on a twelve-month scale after the seasonal adjustment. It does not say that contractors received $2.186 trillion in September, and it does not forecast the value of work that will actually be completed over the next year. The adjustment also means that dividing the headline by twelve is not the same as retrieving the observed, unadjusted September total.
For that different question, the not-seasonally-adjusted table reports $196,349 million in September construction spending. August was $204,944 million, and September 2025 was $202,041 million. Comparing September with August without seasonal adjustment mixes changes in the underlying market with normal seasonal patterns. Comparing the adjusted annual rates is designed to reduce that problem. Comparing the unadjusted September figures across years answers another question about the same point in the seasonal calendar. Neither series is wrong; the error is to switch measures mid-sentence without saying so.
The year-to-date total provides a third perspective. During January through September, construction spending amounted to $1,649.6 billion, compared with $1,698.6 billion in the corresponding period of 2025, a 2.9% decline. That accumulated total is not an annualized September run rate. It incorporates work over nine months, including projects that began before 2026 and may continue after it. A builder deciding whether a recent month reflects the year so far should keep both measures visible rather than letting one unusually large monthly move stand in for the entire pipeline.
Residential spending is broader than new homes
Private residential construction fell to an $848.777 billion annual rate from $897.995 billion in August. The $49.218 billion annual-rate decline in that category was larger than the decline in total construction, because gains elsewhere offset part of it. Yet the new single-family component was $403.436 billion, almost identical to August's $403.477 billion. New multifamily work was $115.888 billion, compared with $115.814 billion. A statement that new-home construction spending dropped 5.5% would incorrectly attach the change in the wider residential category to a narrower activity.
Subtracting the two new-building categories from the private residential total leaves about $329.453 billion in September, versus $378.704 billion in August. That is a calculation from the published rounded table values, not a separately reported confidence interval. The release notes that residential totals include improvements, and the Census definitions describe owner-occupied remodeling, additions and major replacements separately from new construction. The residual calculation shows why improvements deserve close attention when interpreting this release. It does not establish why households or contractors changed their spending, nor does it reveal whether a particular type of renovation drove the movement.
The distinction is practical. Adding a room, finishing an attic or replacing an entire heating system can fall within construction improvements, while routine maintenance and repairs are excluded from the construction definition. A contractor replacing complete systems and a business handling recurring minor repairs therefore may serve the same neighborhood while falling differently within the statistics. The report also does not measure the number of prospective buyers, the inventory of homes for sale or mortgage applications. Those are separate indicators with their own timing and definitions, and none can be inferred directly from the residential spending headline.
The nine-month comparison is less dramatic, but still weaker
The January-to-September private residential total was $671.325 billion, down 4.5% from $703.162 billion a year earlier. Within that, new single-family construction was $306.075 billion, down 4.8% from $321.620 billion. New multifamily spending was $87.141 billion, only 0.3% below $87.385 billion. These comparisons show a softer year for single-family construction spending even though its September monthly rate barely changed. The monthly and cumulative readings are compatible: a stable latest month does not erase earlier weakness.
The distinction also prevents a misleading recovery narrative. A category can stop falling for one month while remaining below its year-earlier level, just as a monthly increase can coexist with a lower cumulative total. September's annualized private residential rate was 8.9% below September 2025 in the detailed table. New single-family spending was 2.4% below its year-earlier rate, while multifamily spending was 0.2% lower. That split is more informative than describing all housing construction as uniformly weak or uniformly recovering.
Dollar spending still is not a physical output count. An expensive project contributes more dollars than a less costly one even if each adds one building. Changes in materials, labor costs, project mix or the stage of work can affect the spending measure without producing a matching movement in unit counts. Because these estimates are not adjusted for price changes, the report alone cannot say how much inflation-adjusted construction activity grew or shrank. A conclusion about physical building volume needs another measurement step, not just a different adjective applied to the dollar total.
Offices strengthened while factory construction remained below last year
Private nonresidential construction rose 1.1% in September to an annual rate of $790.554 billion from $782.272 billion. The office category accounted for a large part of the arithmetic increase: its rate rose from $138.103 billion to $145.633 billion, a gain of $7.530 billion. Total private nonresidential work increased by $8.282 billion. This comparison identifies a major contributor to the net change, but it is not a claim that all other categories were flat. Some rose and others declined, leaving a smaller net contribution after offsets.
Office construction in Census terminology is not restricted to the familiar leased downtown tower. The classification includes computer centers and a data-center subcategory. That matters when readers compare the construction report with anecdotes about empty offices or changes in work-from-home policies. A broader construction category can rise even when one recognizable segment faces difficulty. The headline office total cannot by itself establish how much of September's growth came from any one subtype, and this article does not assign an unverified data-center figure to the increase.
Manufacturing construction showed a different pattern. Private manufacturing work was at a $167.226 billion annual rate, down slightly from $167.675 billion in August and 17.9% below its September 2025 rate. Its first-nine-month total was $128.912 billion, compared with $164.739 billion a year earlier, a 21.7% decline. That is a construction-spending comparison, not a reading of factory production or orders. A newly finished factory can shift from construction activity to operating output, while an operating plant can produce more without undertaking a major building project. The spending series should not be relabeled as a manufacturing-output index.
Power and commercial buildings show why the category map matters
Private power construction rose to a $171.658 billion annual rate from $169.698 billion in August, an increase of 1.2%, and was 12.1% above September 2025. The category covers more than generating stations. Census includes electricity distribution and transmission facilities, as well as specified gas and oil infrastructure. A national power-spending figure therefore cannot be read as the amount spent only on renewable generation, nuclear facilities or a single grid technology. Identifying the project types behind the change requires more detailed evidence than the broad table provides.
Private commercial construction was $115.924 billion at an annual rate, down 0.3% from August and 4.4% from a year earlier. The commercial classification includes retail and wholesale facilities and selected services, with distinctions for warehouses and other uses. The company owning a building does not automatically determine its statistical category. For example, Census classifies an office at a manufacturing site differently from an office owned by a manufacturer but located away from that site. Similar-looking buildings can belong to different rows because the definitions follow their use and context.
That classification issue is particularly important when comparing business announcements with the monthly release. A company may announce a large investment package containing land, buildings and production equipment. Census construction spending excludes land acquisition and most production machinery. A hospital building and its clinical equipment are not all treated as the same construction expense; similarly, racks and servers are excluded from the construction definition for data centers. Adding the full value of announced corporate investment packages and expecting them to match a Census construction row would combine unlike measures.
Public construction was steady within the survey's uncertainty
Public construction was estimated at a $546.8 billion annual rate, down 0.1% from August. The reported margin around that monthly change was plus or minus 1.3 percentage points. Because that range includes zero, the release marks the change as not statistically significant. Calling it a clear public-building contraction would give a tiny point estimate more certainty than the source supports. Public educational construction was virtually unchanged at $113.1 billion, and highway construction was $149.2 billion, down 0.3% with a much wider reported margin of plus or minus 3.6 percentage points.
The year-to-date public total was $409.281 billion, 1.1% above the $404.754 billion recorded in the first nine months of 2025. That contrasts with private construction's $1,240.353 billion cumulative total, down 4.1% from $1,293.859 billion. Public work therefore provided a different year-to-date trajectory, even though it could not offset the larger decline in private spending in the national total. These are ownership-based categories during construction, not a simple classification by who ultimately uses a building or whether the project receives attention from a government agency.
The photographs of hospitals, dormitories and infrastructure illustrate why a public program is a collection of long projects rather than one monthly switch. Work proceeds through site preparation, structural construction, systems installation and finishing. A project announcement, contract award, groundbreaking and ribbon cutting describe different stages. The value-put-in-place approach tries to measure work during a period across projects already underway, regardless of the date each began. It is not a tally of that month's public announcements, and a photograph of a completed building does not identify when most of its construction value entered the national series.
What the confidence intervals and revisions do, and do not, say
The total monthly decline of 1.9% carried a reported margin of plus or minus 0.7 percentage points. Unlike the tiny public-sector change, that range does not include zero. The residential decline of 5.5%, with a margin of plus or minus 1.3 points, and the private nonresidential increase of 1.1%, with a margin of plus or minus 0.5 points, also have ranges that stay on one side of zero. That supports distinguishing the directions of those reported changes at the release's stated confidence level, without pretending the point estimates are exact.
Sampling uncertainty is only one source of uncertainty. The release explains that the confidence intervals address sampling variability, while errors in reporting or processing and other nonsampling issues can also affect results. Preliminary estimates are revised as more information becomes available. August's value in this release is explicitly revised, so the September comparison must use that revised number rather than an older headline saved from the previous publication. A spreadsheet that carries an outdated August value forward can manufacture an incorrect monthly change even when the September entry is copied correctly.
Census also cautions that it can take several months to establish a trend, with the time needed differing between total construction and individual categories. That is not a reason to discard the newest release. It is a reason to read it alongside its revision status, the year-to-date comparison and the composition of the movement. The residual residential calculation in this article is an analytical aid, not a substitute for a published sampling interval or a forecast. When the next release updates the component estimates, that calculation should be redone from the same version of the tables.
A more useful reading list for the next release
For a reader following housing, the first comparison is the private residential total alongside new single-family and multifamily spending, rather than treating the total as a new-home series. For a reader following business investment, the office, power and manufacturing rows deserve separate attention because they have sharply different recent patterns and contain different types of assets. For a reader following public services, the relevant question is whether spending on the specific infrastructure category changes beyond ordinary measurement uncertainty, not whether the aggregate moves a tenth of a percentage point.
The next scheduled release covers October and is due December 1. It will provide another observation and revised recent estimates, not a final verdict on every construction market. September's strongest conclusion is narrower and more useful: national nominal construction spending weakened, a substantial residential movement was not matched by new-house or new-apartment spending, selected private nonresidential categories strengthened, and public spending was broadly steady on the month. Those findings can guide what to examine next without turning one release into a claim about every contractor's order book.
Construction is both an immediate source of work and a long process of adding or changing buildings and infrastructure. The Census tables capture part of that process in dollars. Reading their units, ownership rules and components carefully makes the numbers more concrete, not less. A national decline does not mean every site has stopped, and a rising category does not mean every announced investment has already become work in place. September's report rewards that distinction: the details explain much more than the headline can carry.
Comments
Post a Comment