US manufacturing construction spending has become the consensus shorthand for industrial renewal. The construction data through July 2026 argues the renewal never arrived, and what replaced it tells a more uncomfortable story than the tariff debate usually entertains.
The Numbers the Reshoring Narrative Skips
Private construction spending on manufacturing stood at $167,805 million in July 2026, down 32.6% over two years. That is not a soft patch. The manufacturing boom that began in 2021 peaked in September 2024 and has been declining steeply since. According to IoT Analytics, US manufacturing construction spending has fallen 21% since reaching $239 billion in June 2024, a figure that underscores just how rapid the unwind has been. The computer, electronic and electrical sector, which represented the most technologically intensive slice of the boom, has fared worse still: IoT Analytics puts that segment’s decline at 44% since its own peak in July 2024.
The consensus reads tariffs as a reasonable, if blunt, tool for reversing this trajectory. The data suggests the trajectory was already in motion before tariffs could have had any structural effect, and the direction has not changed.
Year-to-date spending through May declined 22.4% compared with the same period in 2025, according to Engineering Inc (ACEC). That is an acceleration of the retreat, not a stabilisation.
What Is Actually Growing: Data Centres and Power
The headline figure for total private nonresidential construction spending, $238,888 million in July 2026, is up 6.4% over two years. That number has masked what is happening underneath it. Data centres and power infrastructure have effectively taken the place manufacturing occupied in the spending mix. Combined spending on data centres plus power reached $236,707 million by July 2026, against manufacturing’s $167,805 million. In September 2025 the crossover happened: data centres plus power eclipsed manufacturing for the first time.
The structural substitution is visible in the sub-components. Power spending alone stands at $161,541 million, closing in on manufacturing as a single line item. Data centres at $75,166 million complete the picture. The downward slope of manufacturing spending and the upward arc of data centres plus power are, as MishTalk put it, nearly mirror images of each other.
Commercial construction is not compensating either. At $116,241 million, it is down 23.0% over three years. The one sector sustaining aggregate construction spending is the one no politician is campaigning on.
It is also worth noting what this infrastructure build is not generating. Overall US construction spending declined 1.4% in 2025 compared with the prior year, according to ABC Carolinas, meaning data centre growth has not been sufficient to lift the broader market. The fastest-growing construction segment cannot compensate for the weakness elsewhere.
The Jobs Argument Is Weaker Than It Looks
The political case for manufacturing revival rests partly on employment. Here the historical comparisons are instructive and, for the revival thesis, not encouraging. Manufacturing’s share of nonfarm payrolls has followed a long structural decline, comparable in trajectory to farm jobs, department stores, 8-track tapes and cassettes. The pattern is one of permanent displacement, not cyclical dip.
The data centre substitution compounds the problem. A large data centre is a construction-intensive project during the build phase, but its post-construction job creation is limited. Manufacturing facilities, by contrast, generate far denser ongoing employment and broader supply chains. The composition of what is being built matters as much as the aggregate dollar figure, and on composition the current trend is moving in the wrong direction for employment.
The consensus may be overweighting tariffs as a mechanism for reversal while underweighting the structural forces that drove the original decline. The manufacturing boom from 2021 to late 2024 was real; so is its unwinding. US manufacturing construction spending in July 2026 is where it is not because of any single policy choice, but because capital is finding higher returns in digital infrastructure. That calculation does not change because tariffs make imports more expensive. It changes when the underlying economics of building factories shifts, and on that question the spending data through July 2026 offers no sign of a turn.
