Real US manufacturing construction spending more than doubled after 2021 and has since fallen by over a third

Research question

How has real US private spending on manufacturing construction changed since 2021, and how is it distributed across subsectors and regions?

Construction at manufacturing sites is an early physical trace of new industrial capacity. The Census Bureau measures it monthly as the value of construction put in place: the value of buildings and structures installed at manufacturing sites in the period. The measure excludes production machinery, so it tracks plants being built rather than total investment in them. This brief measures the size of the rise after 2021 in constant dollars, which subsectors account for it, and where in the country it took place.

Findings

• Real spending peaked in November 2024 at 2.34 times its 2021 average¹. The peak was US$190.4 billion at a seasonally adjusted annual rate, in constant 2021 dollars¹, against an average of US$81.5 billion in 2021¹.
• The rise is larger in current dollars. The nominal series peaked in September 2024 at US$249.1 billion¹, 3.05 times its 2021 average¹.
• Spending has fallen since. In August 2026, the latest month and still preliminary, real spending was 37% below its peak¹ at US$119.9 billion¹. That is 1.47 times the 2021 average¹.
• One subsector accounts for most of both movements. Computer, electronic and electrical plants account for 68% of the increase in annual spending in current dollars between 2021 and 2024² and for 85.6% of the fall between 2024 and 2025³.
• The Northeast is the only Census region where spending fell, from US$10.7 billion in 2021⁴ to US$6.1 billion in 2025⁵.
• One expectation did not hold. The plan expected every region other than the Northeast to gain share. The Midwest's share was 21.5% in both 2021 and 2025⁶.

Evidence

The first exhibit shows the monthly series in current and constant dollars. The gap between the two lines after 2021 is the rise in construction costs as measured by the producer price index for new industrial buildings.

Annual spending in current dollars rose from US$81.6 billion in 2021² to US$243.4 billion in 2024², and then fell by 12.1% in 2025³. Most of that path belongs to one subsector.

Construction of computer, electronic and electrical plants rose from US$12.1 billion in 2021² to US$122.1 billion in 2024², about 10 times as much². Its share of all manufacturing construction went from 14.8%⁷ to 50.2%⁷. In 2021 the largest published subsector was chemicals, with 36.7% of the total⁷.

The geography shifted over the same period. The West's share of spending rose from 16.8% in 2021⁶ to 25.1% in 2025⁶, and the South accounted for 50.5% in 2025⁶. The Northeast's share fell from 13.1%⁶ to 2.9%⁶.

Within regions the movement was uneven. Spending in the South Atlantic division in 2025 was 5.6 times its 2021 level⁶, and in the Mountain division 4.8 times⁶. Spending in the West South Central division also rose, but its share of the national total fell by 8.6 percentage points⁶. The table gives every region and division, with the relative standard error of each 2025 estimate.

Region or division	2021	2025	Share 2021	Share 2025	Change	RSE 2025
United States	81.6	213.8	100.0	100.0	+0.0	0.7
South	39.7	108.0	48.6	50.5	+1.9	0.8
West	13.8	53.8	16.8	25.1	+8.3	1.3
Midwest	17.6	46.0	21.5	21.5	+0.0	2.0
Northeast	10.7	6.1	13.1	2.9	−10.2	6.5
Mountain	9.9	47.4	12.1	22.2	+10.1	1.2
West South Central	24.7	46.5	30.3	21.7	−8.6	1.1
South Atlantic	7.9	43.9	9.7	20.5	+10.8	1.7
East North Central	11.8	33.3	14.5	15.6	+1.1	2.0
East South Central	7.0	17.6	8.6	8.2	−0.4	2.0
West North Central	5.8	12.7	7.1	5.9	−1.2	4.0
Pacific	3.9	6.4	4.7	3.0	−1.7	8.1
Middle Atlantic	6.9	4.1	8.4	1.9	−6.5	6.9
New England	3.8	2.0	4.6	0.9	−3.7	13.4

Interpretation

The rise in manufacturing construction after 2021 and its fall after 2024 are both accounted for mainly by one subsector⁸. Since 2023 the aggregate series has largely described that subsector's building cycle. Other subsectors grew too, but by far less.

About a third of the nominal rise to the peak reflects higher construction costs rather than more construction, as measured by the price index used here and splitting the rise additively⁹. Statements about the scale of the build-out differ materially depending on which series is quoted.

Even after the decline, real spending remains well above its 2021 level¹⁰.

The data do not say why any of this happened. The series records where and in which subsector construction took place. It carries no information on financing, policy, demand, or project decisions, and this brief does not attribute the pattern to any cause. A fall in value put in place is also what the completion of large projects looks like in this series, as well as what cancellations and delays look like. The two cannot be told apart from these data.

Implications

These follow from the measurements above and are interpretation, not findings.

Anyone citing manufacturing construction spending should state whether the figure is in current or constant dollars. The two give different answers to how large the rise was.

The national total is a poor guide to conditions outside the computer, electronic and electrical subsector, and to conditions in individual regions. Readers concerned with a particular industry or region should use the subsector and divisional series rather than the aggregate.

Construction value is not plant investment. Production equipment is excluded from this measure. Whether equipment spending follows the same path is a separate question that these sources cannot answer.

The most recent months are provisional. The August 2026 figure is preliminary; June has been revised twice and July once, and all remain subject to annual revision. The latest months of the price index are also preliminary. The size of the decline from the peak should be re-read after each release.

Methodology

Data. Three public datasets, retrieved on 13 October 2026. The U.S. Census Bureau's monthly Value of Private Construction Put in Place, by type of construction, at a seasonally adjusted annual rate and not seasonally adjusted, January 1993 to August 2026 (release of 1 October 2026). The Census Bureau's annual Value of Private Nonresidential Construction Put in Place by Geographic Division, 1993 to 2025. The U.S. Bureau of Labor Statistics producer price index for new industrial building construction, series PCU236211236211, not seasonally adjusted, June 2007 to August 2026.

Definitions. Value put in place is the value of construction installed or erected at the site during the period. It includes new buildings and structures, additions, alterations, renovations, and major replacements, and the Census Bureau adds architectural, engineering, and miscellaneous costs. It excludes land, maintenance and repairs, and production machinery and equipment. Manufacturing covers all buildings and structures at manufacturing sites; manufacturers' offices and warehouses elsewhere are classified under other types. Seven manufacturing subsectors are published monthly. "Other manufacturing" in this brief is the total less those seven.

Method. Real spending is nominal spending multiplied by the ratio of the 2021 average price index to the index in the same month, giving constant 2021 dollars. Monthly comparisons use the seasonally adjusted annual rate. The 2021 baseline is the average of that year's 12 monthly real values. Annual figures are sums of the 12 not-seasonally-adjusted monthly values, in current dollars. A subsector's share of a change is its change divided by the change in total manufacturing over the same years. The share of the nominal rise attributed to costs is one minus the ratio of the real rise to the nominal rise, each measured from the 2021 average to its own peak month; a multiplicative split would attribute about a quarter rather than a third. Regional shares are of the national annual total. The annual national total built from the monthly file was compared with the total in the division file; they agree to within US$3 million in every year. The SQL for every figure is kept with project CR-MFG-2026-002.

Assumptions. The producer price index for new industrial buildings is taken to represent the cost of manufacturing construction. It is not specific to any subsector. If the cost of building fabrication plants rose faster than the index, real growth is overstated here; if slower, understated. The index is not seasonally adjusted and is applied to a seasonally adjusted series.

Limitations. For private manufacturing, the Census Bureau benchmarks its estimates to capital-expenditure survey levels for selected years, most recently 2017; estimates for other years are extrapolations from those levels, so the whole period studied here rests on extrapolation. The real peak is part of a plateau: September 2024 was within 1 percent of November in constant dollars, and the ranking depends on month-to-month movement in the price index. Nominal spending was flat from June to August 2026; most of the real decline over those months comes from a rise in the preliminary price index. The estimates come from a sample survey of construction projects, and private nonresidential estimates are raised by 25 percent by the Census Bureau to allow for projects missing from its sampling frame. Regional and divisional estimates carry larger sampling error than national ones: the relative standard error of the 2025 manufacturing estimate is 0.7 percent nationally and 13.4 percent for New England. Relative standard errors are published only for 2025, so the precision of the 2021 regional estimates is unknown; the 2025 relative standard error is 6.5 percent for the Northeast and 8.1 percent for the Pacific division. Sampling errors do not capture coverage or other nonsampling error. Recent months are preliminary or revised, and annual revisions can alter earlier years. Four months of the price index are preliminary. The regional data are annual and are not broken down by subsector, so the subsector and regional findings cannot be combined. The hypothesis about regional shares was written after regional totals had been seen during file inspection and was not a blind test.