July 2026: Construction Spending Is Softening. Data Centers Are Surging as the Factory Boom Rolls Over

Key Takeaways

• Flat in July, clearly down for the year. Total construction ran at a $2,157.6 billion seasonally adjusted annual rate. The 0.5% monthly dip (±0.8%) can’t be distinguished from zero, so July itself was flat — but spending is 3.8% below the same month a year ago (±1.5%) and 3.5% below last year’s January–July pace (±1.0%).
• Cost-adjusted, most of the “boom” is inflation — but cost-adjusted spending did rise, and is now moderating. Deflating the Census spending series by construction-input price indexes — a proxy, not a direct measure of construction output — leaves this year’s January–July total about 8% above the 2016–2019 average for those same months, versus roughly 69% in nominal dollars. So most of the nominal surge is higher costs rather than more building — but not all of it: cost-adjusted spending still sits above its pre-pandemic average, having climbed to about +19% by 2024 and eased to +8% since.
• The AI-adjacent construction cycle is changing shape — and shrinking. Computer, electronic and electrical factory construction is down about 45% this year, while data centers are up roughly 35%. But data-center growth has replaced only about one-third of the factory decline.

What happened

Construction spending slipped again in July, but the monthly move is too small to call. At a $2,157.6 billion annual rate, total spending was down 0.5% — with a ±0.8% margin of error, that’s statistically indistinguishable from zero, so July is best read as flat. The signals that clear the significance bar are the longer comparisons: spending is 3.8% below July 2025 (±1.5%) and 3.5% below last year’s January–July pace (±1.0%).

The weakness is broad but uneven. Every year-to-date figure here compares January–July 2026 with the same seven months of prior years — the same-period comparison that nets out seasonality without extra smoothing. Private construction spending is down 4.8% year-to-date; public spending is essentially flat (+0.8%, ±1.8% — not statistically significant). Within private construction, residential spending is down 4.1% year-to-date (±2.1%) and nonresidential is down 5.6% (±1.0%). On the residential side, single-family put-in-place is down 5.7% while multifamily is roughly flat (−0.3%) — the same single-family-led cooling that shows up in starts and residential construction activity.

None of that is a collapse. It’s a market decelerating from an unusually high base, and — as the rest of this piece argues — the composition underneath is the real story.

The big picture

Adjust for prices and the “record spending” narrative shrinks — and turns into a rise and a moderation. In nominal dollars, this year’s January–July total sits about 69% above the 2016–2019 average for the same months. But construction costs have risen sharply since 2019, and deflating by a construction-input price index cuts that to roughly +8%. It also reveals a peak-and-decline the nominal series hides: cost-adjusted spending climbed to about +19% above the pre-pandemic average by 2024 and has since eased to +8%. Residential shows the same shape — about +70% above its pre-pandemic average in dollars, but a cost-adjusted +10% that peaked near +27% in 2022.

Manufacturing is the biggest drag — by a wide margin. Factory construction is still roughly 49% above its 2016–2019 average on a cost-adjusted basis — but that’s down from a +119% peak in 2024, and spending is now falling fast: down 22.2% year-to-date and 21.2% from a year ago. In dollar terms, factory construction fell about $28.8 billion from last January–July to this one — roughly two-thirds of the $45.1 billion decline in total construction. It doesn’t single-handedly explain the drop (had manufacturing merely held flat, total spending would still be down about 1.3% year-to-date), but it is the largest single contributor by far.

The real declines elsewhere are in the rate-sensitive middle. Commercial construction is down 5.9% from the same January–July period a year earlier and has slipped back just below its 2016–2019 average in cost-adjusted terms; lodging and traditional transportation projects sit well below theirs. These are the categories most exposed to borrowing costs and cautious corporate budgets — where the softness is real, not just nominal.

Public construction provided only a modest offset. In cost-adjusted terms, highway, water, sewage and public-power construction all remain above their 2016–2019 averages — even though each is below last year’s January–July pace. But the arithmetic is small: public spending added about $2.3 billion year-to-date against a $47.5 billion decline in private construction — offsetting only about 5% of the private drop. It cushioned the total; it didn’t hold it up.

Does the “AI is carrying the economy” story show up here? Yes — but as a rotation, not a single engine

The backdrop to almost every 2026 macro conversation is that artificial-intelligence investment is doing an outsized share of the economy’s work. Construction is a good place to test that, because the AI capital cycle takes physical form: the factories that make the chips, the buildings that house the servers, and the power that runs them. What the data show is not a single AI engine but a rotation — from factories toward data centers, even as combined spending across the two categories shrinks.

The first wave — factories — is fading. The semiconductor and electronics plants at the center of the AI supply chain sit inside the manufacturing category. Within private manufacturing, the computer/electronic/electrical subcategory is down about 45% year-to-date — a roughly $28 billion drop that accounts for nearly all of the decline in private factory construction and a large share of the fall in total construction. Two clarifications matter here. First, this category is AI-adjacent and semiconductor-heavy, but it is not synonymous with AI — it includes substantial non-AI industrial investment. Second, some of the decline is mechanical. As big CHIPS-era projects move from heavy construction toward equipment installation, they generate less construction put-in-place spending even as the facilities themselves move closer to production. That means falling construction spending does not necessarily imply that semiconductor investment itself has collapsed — though delays, cancellations, and fewer new megaprojects entering the pipeline could also be at work, and the Census data alone can’t separate those causes. Part of the factory-building boom isn’t failing. It’s aging.

The second wave — data centers — is surging. Traditional office construction has been moribund since remote work reshaped demand, yet total private office spending is up 10.2% year-to-date. Open the category and the reason is unambiguous: Census now breaks out data centers as a distinct component of private office construction (back to 2014). Through July, data-center construction is up about 35% year-to-date while general office is down 11.8% — and data centers make up roughly 57% of all private office construction. Data centers aren’t merely padding the office number; the data-center gain is about 1.6 times the entire net increase in private office spending, meaning conventional offices are shrinking underneath it. The gap is no longer close: by July, data-center construction was running at roughly a $75 billion annual rate, compared with just $45 billion for general office. The “office recovery” is really a data-center infrastructure boom wearing an office label — with AI demand a major force behind it.

Power is growing alongside it — in nominal dollars — consistent with, but not provably caused by, AI. Power construction is up 5.3% from a year ago (±1.2%) and 3.6% year-to-date in nominal dollars. After adjusting for input costs, however, January–July spending is about 2% below last year — though still roughly 12% above its 2016–2019 average, so elevated rather than currently accelerating. That is consistent with the enormous electricity requirements of hyperscale data centers, and it’s plausible that grid and generation investment is following the data-center pipeline — the Energy Information Administration forecasts the strongest four-year growth in U.S. electricity demand since 2000, driven largely by large computing facilities. But the Census construction data don’t tell us how much of today’s power spending is directly attributable to AI demand — power construction has other drivers (broader electrification, grid hardening, renewables), so this remains an association, not a decomposition.

Does AI-adjacent infrastructure show up? Clearly — but the sharper reading is that the cycle is rotating from factories toward data centers while shrinking overall. Computer, electronic and electrical factory construction is rolling over as data-center construction accelerates. Yet data-center growth has replaced only about one-third of the factory decline: combined spending in the two categories fell from roughly $88.7 billion in January–July 2025 to $70.6 billion this year. The center of gravity is moving, but this is not a one-for-one handoff.

Momentum: the year-to-date picture

Looking through the monthly noise at the January–July totals — the same-calendar-month comparison that handles seasonality natively — the story is soft but not alarming. Total spending is down 3.5% from last year; private residential down 4.1%; private nonresidential down 5.6%. The only major grouping in the black is public construction, and only barely.

Measured against a fixed 2016–2019 average, cost-adjusted total construction climbed to about +19% by 2024 and has eased to +8%. That isn’t a plateau, but it isn’t a bust either: it’s a rise and a moderation, still comfortably above the pre-pandemic average, with the extraordinary factory cycle now unwinding on top of it. The pattern repeats sector by sector, and the shapes differ in telling ways: manufacturing towers to +119% and has since dropped to +49%; residential peaks at +27% and eases to +10%; office is the exception — it fell to 12% below its pre-pandemic average in 2022, then recovered to 14% above it as data centers surged. (The exact levels shift a little under a different pre-pandemic window, such as 2018–2019, but the shape doesn’t.)

The bottom line

Construction spending was statistically flat in July but is clearly softer for the year, down 3.5% — dragged first and foremost by a fast-fading factory-construction cycle and a financing-sensitive commercial middle.

Adjust for costs and the picture is calmer than the nominal declines imply: cost-adjusted spending is still about 8% above its 2016–2019 average — a rise and a moderation off a +19% peak in 2024, not a collapse.

And the AI-adjacent construction story is real, with a twist. Computer, electronic and electrical factory construction — the category that powered the industrial boom two years ago — is now the sector’s biggest drag. Data centers are surging in its place, while power construction remains elevated and continues to grow in nominal dollars. The center of gravity is moving from factories toward computing infrastructure, but this is not a one-for-one handoff: data-center growth has replaced only about one-third of the factory decline.

Sources: U.S. Census Bureau, Construction Spending (Value of Construction Put in Place), July 2026 release (CB26-140), issued September 1, 2026, seasonally adjusted and not-seasonally-adjusted series, including the detailed private-construction time series with the data-center breakout within private office construction. Margins of error are 90% confidence intervals from Census Table 3; changes whose intervals include zero are flagged as not statistically significant. Cost-adjusted (“real”) figures deflate the Census spending series by U.S. Bureau of Labor Statistics Producer Price Index construction-input indexes used as a proxy deflator — these are model-derived estimates, not Census figures, and carry no Census margin of error. The pre-pandemic benchmark is each series’ average January–July total over 2016–2019 (a fixed baseline); every year is then expressed as a percent of that average. An alternative window (e.g. 2018–2019) shifts the levels modestly but not the overall rise-and-moderation pattern. Data-center and general-office annual-rate figures are from the current (July 2026) Census vintage, which revised the earlier-2026 history upward. On electricity demand: U.S. Energy Information Administration, which identifies data centers as a principal driver of the recent acceleration in U.S. electricity demand.