NextFin News - The AI data center construction boom is real, but it is not big enough to stop the broader US construction slump. In July 2026, total construction spending fell 0.5% from June and 3.8% from a year earlier, the Census Bureau reported, and the entire month-to-month gain in nonresidential work came from a single category: data centers. Strip that out and nonresidential construction has now contracted for two straight months to its weakest level since September 2023.
The tension is stark. Data center construction spending reached a seasonally adjusted annual rate of $50.7 billion in April 2026, more than all spending on transportation infrastructure — airports, rail terminals, marinas — combined. Yet a boom of that size still cannot offset weakness in a roughly $2.2 trillion construction market where residential activity is down 7.3% year over year and manufacturing construction has collapsed 21.2%.
The Boom in One Number, the Slump in Another
July's Census data reads like two different economies stapled together. Total construction spending slipped 0.5% month over month and 3.8% year over year. Residential led the decline, falling 1.3% in July alone and 7.3% against July 2025. Nonresidential spending was essentially flat at 0.1% month over month but still down 1.3% year over year — and that near-flat reading masks how concentrated the strength is.
Office construction, the Census category that includes data centers, rose 16.9% year over year. The power sector, up 5.3% yearly, benefited from the same buildout as hyperscalers demand electricity at industrial scale. Everything else weakened: manufacturing fell 21.2%, lodging dropped 9.6%, commercial construction declined 4.9%. Only highway and street work, up 4.5%, joined the data center complex on the positive side.
"The increase in nonresidential construction spending that occurred in July was entirely due to data centers," Anirban Basu, chief economist at Associated Builders and Contractors, said in a statement. "Excluding that booming category, nonresidential spending fell for the second straight month and is down to the lowest level since September 2023."
The data center surge is not a rounding error. Through April 2026, data center construction starts totaled $49.5 billion, nearly four times the $13.6 billion recorded in the same period a year earlier, according to ConstructConnect. Average project cost over the trailing twelve months climbed to $475 million, from $178 million a year earlier, as AI-optimized facilities demand denser compute, liquid cooling, and dedicated power. The three-month moving average of monthly starts sits around $8.0 billion.
But context matters. In a total construction market running at roughly $2.17 trillion a year, even a $50.7 billion-plus annualized data center sector is about 2.3% of the whole. When residential — a far larger share of the market — sheds 7.3% year over year and manufacturing sheds more than a fifth of its spending, the math does not close. The boom is deep, but it is narrow.
Why the Rest of Construction Is Falling
The manufacturing collapse deserves its own explanation, because it is the other half of this story. Manufacturing remains the largest nonresidential construction market, yet year-to-date spending through May 2026 fell 22.4% against 2025. This is not a mystery of weak demand; it is a statistical hangover from an extraordinary boom.
The United States spent 2022 through 2024 building semiconductor fabs and battery plants under the CHIPS Act and the Inflation Reduction Act. That wave pulled manufacturing construction to record levels. Once those megaprojects moved from groundbreaking to equipment installation, the year-over-year comparisons turned brutal. Michael Guckes, chief economist at ConstructConnect, put it plainly: manufacturing's decline is "in large part due to tough prior year comparables, especially in the construction of computer and electronics manufacturing structures." His conclusion: "These contractions are being only partially offset by strong gains in data centers and power."
That is the cyclical leg of the story. Semiconductor fab construction is a lumpy, investment-cycle business. When a handful of $20 billion plants finish their shell construction in the same window, the following year's comparisons look catastrophic even if the industry is healthy. The same dynamic shows up in lodging, down 9.6%, and commercial construction, down 4.9% — both cooling after the post-pandemic rebound ran out of steam.
Residential weakness, by contrast, is rate-driven and broad. Housing starts in July 2026 fell 12.4% from June to a seasonally adjusted annual rate of 1.239 million units, near a six-year low. Single-family starts dropped 9.9% and multifamily starts sank 15.6%. With mortgage rates still above 6%, affordability remains the binding constraint, and builders have pulled back from the multifamily segment after working through record apartment backlogs. The National Association of Home Builders expects multifamily starts to fall 5% in 2026.
So the slump has two distinct engines: a cyclical manufacturing hangover and a rate-sensitive residential slowdown. Data centers sit outside both. They are not a substitute for homebuilding, and they do not hire the same trades in the same places at the same scale.
The Structural Shift Underneath the Cycle
Here is where the story turns. The data center boom is not a cyclical sugar rush; it is a structural shift in where construction capital is going. Artificial intelligence has turned data centers from a niche industrial use into the dominant destination for US construction dollars in the private sector. Year-to-date data center spending through April 2026 was roughly 3.6 times the prior-year pace. Full-year 2025 data center construction starts totaled $77.7 billion, up 190% from 2024's $26.9 billion, which itself was up from $14.9 billion in 2023 — a four-year compound annual growth rate near 98%, according to industry data.
The capital commitments backing this are multi-year and corporate, not rate-sensitive. Five major technology companies have announced roughly $700 billion in capital expenditure for 2026, with Amazon, Microsoft, Google, and Meta each trending toward $100 billion annually. Goldman Sachs estimates baseline AI-related capital spending of about $765 billion in 2026, rising to $1.6 trillion a year by 2031. These are not projects that get cancelled because the Federal Reserve holds rates steady; they are the physical infrastructure of the AI race, and their timeline is measured in chip generations, not business cycles.
The structural footprint extends beyond the server halls. Every data center needs power, and the grid is now part of the construction story. Power construction spending is up 5.3% year over year, and transmission and distribution work follows the campuses. The near-term pipeline is heavily concentrated in the South, which accounts for well over half of planned data center spending, according to ConstructConnect. That is a durable, multi-year demand stream for electrical contractors, substations, and generation — a structural tailwind embedded in the industrial base.
But a structural shift in one sector is not the same as a structural boom in construction overall. The transmission mechanism from AI capex to the broader construction labor market is constrained by three bottlenecks. First, geography: data center activity is heavily concentrated in a handful of states while residential and manufacturing weakness is national. Second, specialization: the trades in shortest supply are electricians, MEP engineers, and commissioning specialists, not the framing and finish crews that build homes. Third, capacity: Associated Builders and Contractors estimates the industry must attract 349,000 net new workers in 2026 and 456,000 in 2027 just to meet existing demand, with about one in five electricians already over 55.
The irony is that the data center boom is itself a symptom of the constraint. Because labor and power are scarce, projects take longer and cost more — the average data center project now costs $475 million, up from $178 million a year earlier — which means each dollar of capex buys fewer completed structures and employs fewer workers per dollar than it did. The boom is measured in dollars, not in breadth.
The Counter-Argument: This Time the Boom Could Broaden
The bull case deserves a full hearing, because it attacks the core thesis at its foundation. The argument is simple: the data center complex is not a single narrow category. It pulls in power generation, transmission, communications infrastructure, and industrial construction. With hyperscaler capex commitments stretching into 2031 and data center spending already exceeding all transportation infrastructure, the boom could be the seed of a broader industrial renaissance rather than an isolated spike.
There is evidence for this. Office construction including data centers is up 16.9% year over year. Power is up 5.3%. The three-month moving average of data center starts sits at $8.0 billion a month, and the pipeline of announced projects — 548 data centers valued at $513 billion between 2026 and 2028, according to Industrial Info Resources — suggests the spending has years to run. If that capital spills into grid modernization, semiconductor-adjacent industrial work, and the supply chain of electrical equipment, the "narrow boom" diagnosis could age badly.
The answer is that spillover is real but insufficient at current magnitudes. Data center spending alone, at a $50 billion-plus annualized rate, is about 2.3% of the total construction market. The rest of private nonresidential construction — a base of roughly $745 billion — is contracting, and residential construction, a separate $877 billion market, is down 7.3% year over year. A 16.9% gain in the office category cannot mathematically offset a 21.2% collapse in manufacturing and a 7.3% decline in residential unless the boom grows several times larger, or the rest of the market stabilizes on its own.
The falsifying signal is specific: if nonresidential construction excluding office and power turns positive for three consecutive months, and manufacturing construction stabilizes above prior-year levels, then the "too narrow" thesis is wrong and the AI boom has become broad-based. Until that prints, the concentration risk stands.
What Comes Next: Three Horizons
Short term (sentiment and liquidity): The next few months of Census releases will show whether July's pattern repeats. Watch the month-to-month change in total construction spending and the nonresidential reading excluding data centers. A second or third consecutive monthly decline outside data centers would confirm the slump is self-sustaining, not a one-month blip. Housing starts and building permits are the swing factor — a rebound there would cushion the headline.
Medium term (fundamentals): The manufacturing comparison base should ease through 2027 as the semiconductor fab wave fully rolls off, which should mechanically lift the year-over-year manufacturing number even without new strength. That is the most likely path to stabilization: not a data-center-led boom, but a cyclical bottom in factories and homes. Beneficiaries of that path are the trades and materials suppliers exposed to broad nonresidential work — commercial contractors, lodging builders, and manufacturers of building products — rather than the specialized electrical and commissioning firms riding the data center wave.
Long term (structural): The AI buildout is a multi-year structural shift in the composition of construction demand. Power infrastructure, electrical contractors, and regions hosting hyperscale campuses should outperform the aggregate market for years. But the aggregate market is still driven by housing affordability, interest rates, and the manufacturing investment cycle — none of which AI resolves. The exposed parties are homebuilders tied to rate-sensitive first-time buyers, lodging developers in markets that overbuilt during the pandemic rebound, and commercial contractors with little data center or power exposure.
Base case: total construction spending remains flat to slightly negative through late 2026, with data centers and power offsetting only part of the residential and manufacturing drag. Upside case: mortgage rates fall toward 6%, permitting rebounds, and the manufacturing comparison base normalizes faster than expected, lifting total spending back to growth in early 2027. Downside case: rates stay higher for longer, housing starts break below the six-year low, and data center project delays from labor and power constraints slow the one category holding the market up.
The AI data center boom is the strongest construction story of the decade, and it still cannot fill the hole left by homes and factories. That is not a bet against artificial intelligence; it is a reminder that a $50 billion boom in a $2.2 trillion market is a sector rotation, not a cycle. The market that matters is not how fast data centers are rising, but whether everything else stops falling — and on that question, the data center boom has no answer.
Data as of the July 2026 US Census Bureau construction spending release, published August 2026.
Explore more exclusive insights at nextfin.ai.
