Data centers are being built faster than estimators are being hired. Spend is up 79% in two years, the estimator workforce is shrinking, and only one contractor in eight has any of the work. The rest are not losing on price. They are never bidding at all, because nobody on the desk has the hours. This report shows the gap, and what closes it.
Sources: US Census Bureau · Associated Builders and Contractors · our survey of 312 US industrial estimators, published as The State of Industrial Estimation 2026 and The Shortage of Estimators. Full sources in the appendix.
White cards are third-party figures reported as published. Dark cards are our own survey or our own arithmetic from public data. The two are never blended in a single number. Methodology and sources at the foot of the page.
Spend rose 79% in two years. The estimator workforce is projected to shrink. When a desk cannot get through a package on the deadline, the contractor passes, the work stays with the 12% who already have it, and the boom shows up in someone else’s backlog. theTakeoff.ai exists to change which side of that line a desk sits on.
The Census Bureau broke data centers out as their own category this year. The first clean reading put them past general office construction and past every airport, port and transit project in the country combined.
The first clean reading came in April 2026: $50.7 billion at a seasonally adjusted annual rate (SAAR), which is a monthly figure scaled to what it would be over a full year.
That number passed general office construction ($43.8 billion) and the whole of public transportation construction, meaning airports, ports and mass transit combined, for the first time. Data centers are now 2.3% of every construction dollar spent in the United States.
April 2024 was $28.3 billion. April 2025 was $39.6 billion. April 2026 was $50.7 billion. No other private nonresidential category moved like that. Manufacturing, the previous boom, fell 22% over the same year as CHIPS Act megaprojects moved from civil work into equipment fit-out.
Spend counts work already under way. The better signal for an estimating desk is what broke ground recently, because the packages behind those projects are still being bid out to the trades.
Every one of those projects was priced months before the first pour. ConstructConnect’s June 2026 Data Center Report puts data center starts through April at $49.5 billion. The same four months of 2025 came to $13.6 billion. Full-year 2025 starts were $77.7 billion, up 190% on 2024, and 2026 is running ahead of that pace.
CBRE’s midyear count tells the same story from the other end. Capacity under construction across the eight primary North American markets hit a record 7,481 megawatts in the first half of 2026, up 24.8% in six months. More than 80% of it is already leased, and vacancy sits at 1.4%. The work is sold before it is built.
It is landing on the corridor that already builds and maintains every refinery, LNG terminal and petrochemical plant on the Gulf Coast, drawing from the same pool of pipefitters, welders and estimators.
Texas leads the country with 26 gigawatts of data center capacity under construction, per JLL data reported by the Houston Business Journal on 20 August 2026. Virginia, the market everyone still calls the largest, has 13. CBRE expects West Texas alone to be a top-five colocation market by 2028.
Louisiana is next door. Meta’s Hyperion campus in Richland Parish has grown from a $10 billion announcement to a $27 billion program, built by Turner, DPR and Mortenson, with peak onsite employment expected above 5,000 workers. Entergy will build ten gas-fired plants totaling 7.5 gigawatts to power it, a 30% increase to the whole state’s grid. Every one of those plants is a piping and mechanical job.
BIC Magazine put it plainly in July: gigawatt-scale digital infrastructure is drawing from the same pool of pipefitters, electricians and welders the industrial sector has relied on for decades. The data center industry alone is projected to need 140,000 more of them by 2030.
One caveat belongs here. Cleanview’s April tracker counts 140 planned Texas projects at 75,089 megawatts, and both Cleanview and ERCOT say a large share of that is speculative. Texas also paused new approvals on 3 August pending an audit. Treat the pipeline as an upper bound. Even a quarter of it is a decade of mechanical and electrical work.
Here is the number that decides whether this boom reaches your desk. Only 12% of contractors are under contract on a data center, and they are full for nearly a year.
ABC’s July 2026 Construction Backlog Indicator found that only 12% of contractors are under contract on a data center. They carry 11.4 months of backlog. The other 88% carry 7.5, and backlog fell in every region, sector and company size that month. The South was the only region with more work than a year ago.
ABC Chief Economist Anirban Basu said the data center boom masks the weakness everywhere else, and that the gap has hit small and mid-size contractors hardest. Firms in the $30 to $50 million revenue band have their thinnest backlog since March 2020. Earlier in the year, fewer than 6% of small firms reported data center work, against 37% of the largest.
The concentration is a structural fact about who bids on hyperscale packages, not a comment on anyone’s estimating. But it changes what the boom means. For most contractors the question is how to get into this work, and the answer runs through the desk: the mechanical, electrical, civil and specialty scopes that spill out of every campus are bid by the trades, one package at a time, on the same deadlines as everything else.
A data center is a power and cooling building. Liquid cooling moves a third of its cost into mechanical scope, and mechanical scope is process piping.
Turner & Townsend’s cost index puts electrical at 54% of an air-cooled facility’s cost and mechanical at 22%. Move to the liquid-cooled halls that AI racks need and mechanical rises to 33%, with electrical at 48%. JLL expects 80% of new facilities to be liquid-cooled by 2030.
That mechanical third is chilled water, coolant distribution loops, secondary pumping, dry coolers, generator fuel and fire suppression. In estimating terms it is process piping, and process piping is the heaviest scope in our survey of 312 industrial estimators: 38 estimator-hours for a mid-size package, more than any other trade.
The package is also getting more expensive to get wrong. JLL puts shell-and-core cost at $10.7 million per megawatt in 2025, up from $7.7 million in 2020, with a further 6% rise expected this year. AI-optimized halls run past $20 million per megawatt. Equipment lead times average 42 weeks, and 57% of 2025 projects slipped at least three months.
theTakeoff.ai reads the piping and instrumentation diagrams, isometrics and drawing sets in a package and returns a draft takeoff, every line, fitting and valve traced to the sheet it came from. The estimator checks a first pass instead of building one, then prices it.
That is the difference between a desk that can take the data center package this month and a desk that has to pass on it.
BOOK A WALKTHROUGH WITH YOUR PACKAGE →Costs are moving under the estimate while it sits on the desk. Input costs rose twice as fast as bid prices over the past year.
AGC’s August 2026 analysis of BLS producer price data shows construction input costs up 7.1% year over year while bid prices rose 3.5%. Nearly half of Turner & Townsend’s respondents reported bid or tender price increases of 6% to 15% in the past year, and another 21% saw more than 15%.
AGC’s outlook found 70% of contractors were affected by tariffs in 2025. Forty percent raised bid prices, a third pulled materials purchases forward, and 11% absorbed most or all of the cost themselves. A number that was right when the package arrived can be wrong by the time it is submitted.
At $10.7 million per megawatt, a contingency added to cover what a rushed team could not see is worth millions on a single hall. It is rational protection, and it is also the reason a thinner desk loses to the desk that could count everything.
BLS counts 221,400 cost estimators in the United States and projects fewer by 2034. Every opening through the decade is a replacement, none of them growth.
BLS counts 221,400 cost estimators as of 2024 and projects 212,100 by 2034. Over the same decade it expects 16,900 openings a year, every one of them a replacement for someone retiring or leaving the profession. Run the arithmetic and 76% of today’s seats turn over by 2034. We took that apart in full in The Shortage of Estimators.
The wider trades are in the same position. ABC’s January model says construction must attract 349,000 net new workers in 2026 to keep up, rising to 456,000 in 2027. AGC’s outlook found 82% of firms struggling to fill craft roles and 80% struggling to fill salaried ones, the highest share in three years, and 60% had postponed or cancelled a project in the past six months because of staffing. McKinsey estimates 20 skilled trade openings for every new entrant through 2032.
Our survey shows what that looks like inside an estimating team. Bid volume per team is up 24% over five years. Estimator headcount is up 4%. Top-quartile desks price 68% of the work they qualify; the median desk prices 45%. A desk that absorbed a quarter more volume on the same people did not have a free quarter to change how it prices.
Hiring will not close this. There are 16,900 openings a year and a shrinking pool to fill them from. The desks that get into the data center work will be the ones that get more packages out of the estimators they already have.
Most coverage of the boom stops at power, land, permits and craft labor. Before any of those constraints bite, someone has to price the job accurately enough to pursue it.
A May 2026 piece in For Construction Pros made the point this report is built on: the preconstruction teams doing that pricing have not scaled with the volume.
The mechanism compounds. A contractor who cannot price a complex package on the deadline passes on it. Every firm that walks away removes price tension from the market. The bids that do go in carry wider margins to cover what the team could not see, and at data center values that padding is structural inflation. Incomplete takeoffs then generate rework, revise, clarify and rebid, before a foundation is poured.
Contractors know where the leverage is. AGC’s outlook found 61% of firms using or planning to invest more in AI, up from 44% a year earlier, and estimating was one of the two most common places they put it, alongside preconstruction. Twenty-three percent already use it for estimating.
Put the curves together. Spend rose 79% in two years, the profession that prices it is projected to shrink 4%, and the gap is 80 index points wide.
Data center construction spend rose 79% in two years. The estimator workforce is projected to shrink 4% over the coming decade, with three quarters of the seats changing hands along the way. The contractors with data center work are full for nearly a year. Everyone else is bidding more to stand still.
The gap will not close by hiring, because there is nobody to hire. The only variable a contractor controls this year is how many hours a package takes to count, and on a data center package the counting is mostly pipe.
That is exactly the variable theTakeoff.ai moves. A draft takeoff comes back with every quantity traced to its sheet, the estimator checks it rather than building it, and the same desk gets through more packages in the same week. More packages priced is more data center work bid, and more work bid is the only way onto the 11.4-month side of the line.
The next data center or industrial package that reaches your desk, log two numbers: the estimator-hours the mechanical and piping takeoff took, and whether the desk could take the same package again next month without dropping something else.
Two packages in, you have the number this report cannot give you, in the unit you control. The Your Missing Millions calculator turns it into hours, packages and dollars against the 312-desk benchmark, no signup.
| Measure | Where it stands | Direction |
|---|---|---|
| Data center construction spend | $50.7B SAAR, +79% in two years | Rising |
| Contractors under contract on a data center | 12%, carrying 11.4 months of backlog | Concentrating |
| Mechanical share of a liquid-cooled hall | 33%, up from 22% air-cooled | Rising |
| Cost estimators employed | 221,400, to 212,100 by 2034 | Falling |
| Estimator seat turnover by 2034 | 76% (calculated) | Compounding |
| Qualified work priced, top quartile vs median | 68% vs 45% | The gap this report is about |
Every contractor we talk to on the Gulf Coast has the same two problems in the same week. A data center package they cannot staff and a refinery turnaround they cannot afford to lose. The estimator is the same person in both meetings. We built theTakeoff.ai so that person gets through the counting in hours and spends the week on the judgment nobody else in the building can do.
theTakeoff.ai reads piping and instrumentation diagrams, isometrics and drawing sets and produces a draft takeoff with every quantity traceable to its source sheet. The estimator reviews it, corrects it and prices it. The 38 hours become a review, and the package you would have passed on gets bid.
Bring that package. It is the one where the difference shows.
The 76% turnover figure is our calculation: 16,900 annual openings × 10 years ÷ 221,400 employed in 2024. Exhibit 13 indexes Census spend and the BLS decade projection to 2024; the headcount points are interpolated. Third-party figures are reported as published; publication dates vary.
theTakeoff.ai builds AI takeoff software for industrial piping and mechanical contractors. The engine reads P&IDs and drawing sets and produces structured, auditable draft takeoffs with click-to-verify traceability to the source drawing and the governing standard. theTakeoff.ai is a product of ContraVault AI, whose bid-intelligence platform is used by EPC and infrastructure contractors including Kalpataru, Rithwik, and ISGEC.
