On April 1, the U.S. Department of Labor announced a national initiative to integrate artificial intelligence skills into Registered Apprenticeship programs. Three weeks earlier, North America's Building Trades Unions and OpenAI announced a partnership to train construction workers for the AI economy. Sam Altman stood next to NABTU President Sean McGarvey and promised that AI's benefits would "reach everyone," starting with "the workers who will build the infrastructure that powers it."
Two announcements in three weeks. AI curricula, modernized training, workforce pipelines for the future. The language was polished. The photo ops were good.
Nobody mentioned that the industry loses roughly one in five of its workers every single year.
The Treadmill That Eats Your Workforce
Bridgit, a workforce planning platform used by nearly 40% of the ENR 400, published its 2026 Construction Workforce Benchmark Report on March 31. The dataset covers 233 contractors and more than 114,000 workers. What it found should alarm anyone spending money on training programs of any kind.
The median attrition rate across the industry is 18.7%. The average is worse: 20.7%. That means a contractor who needs to add 100 workers has to hire approximately 125, because a fifth of the existing workforce is walking out the door while new hires walk in. At the higher end of the distribution, where attrition hits 35%, you need 154 hires to net 100.
Bridgit calls this the "treadmill effect." In 2025, 71.7% of contractors grew their headcount. But 46% of all contractors achieved zero net growth. The hiring happened. The staying didn't.
If you're a residential GC running a 30-person crew and your attrition rate is 20%, you lose six workers a year. You hire eight to replace them and add two. By the time your new AI curriculum finishes its first module, two of those eight are already gone.
Apprenticeships: Half In, Half Out
The Department of Labor's own apprenticeship completion data paints a consistent picture across a decade of records (FY 2013 through FY 2024). In construction trades, completion rates for registered apprenticeships have historically hovered near 50%, varying by trade and state. Electrical programs tend to run higher. Laborers and general construction trend lower.
A four-year electrical apprenticeship starts with a cohort. By year two, you've lost a quarter. By graduation, you're lucky to keep half. Reasons vary by trade and geography, but the Minneapolis Federal Reserve surveyed the problem in 2023 and found structural constraints: programs can't grow faster than the number of qualified mentors, facilities, and willing employers. And even growing programs aren't keeping pace with demand.
Two in three construction firms that are actively hiring say labor availability is their top challenge, according to the Minneapolis Fed's industry survey. Not skills. Not technology gaps. Availability. Warm bodies who show up and stay.
Run the Math on AI-Trained Survivors
Here is a calculation nobody in Washington or Palo Alto offered during their announcements.
Start with 100 apprentices enrolled in a new AI-augmented training program. Apply the national completion rate: 50 finish. Those 50 enter the workforce. Apply the median first-year attrition rate: 9 or 10 leave within 12 months. You're left with roughly 40 AI-trained workers from 100 program slots.
NABTU invests $2.5 billion annually across 1,900 training facilities for its 3 million members. That works out to about $833 per worker per year in training overhead. If an AI curriculum adds even 10% to per-apprentice costs, and 60% of the investment walks away before contributing a single AI-informed day on a job site, the cost per surviving AI-trained worker doubles.
For a 50-person residential builder paying into a union training fund, the question isn't whether AI skills are valuable. It's whether AI training money would produce more returns than spending that same money on the reasons people leave.
Why People Actually Leave
Bridgit's data reveals something the AI announcements carefully avoid. Senior project managers have an attrition rate of 3.6%. Non-senior project managers: 18.8%. The gap is fivefold.
People don't leave because they lack AI skills. They leave because the job is physically punishing, the hours are unpredictable, the pay during apprenticeship is low relative to alternatives, and the path from junior to senior is long and poorly defined.
The Bureau of Labor Statistics reports the median journeyman plumber wage at $30.27 per hour. Apprentices start at roughly half that. A second-year plumbing apprentice earns $16 to $20 per hour, depending on the local and the year of training. Meanwhile, Amazon raised its average warehouse worker pay to more than $23 per hour in 2025, with $5-per-week health insurance, a free Prime membership, and shifts that end at a predictable time. No crawl space work in January. The apprentice's career ceiling is higher. But ceilings don't pay this month's rent, and the warehouse is hiring today.
ABC's 2026 workforce analysis estimates the industry needs 350,000 new workers this year, down from roughly 500,000 in prior years. That drop isn't because demand fell. Construction spending hit record levels. The estimate declined because analysts adjusted for political disruptions to immigration policy and recognized that some of the gap is structural, baked into an industry that has run short-handed for a decade.
What AI Training Could Actually Do (and What It Can't)
Honest accounting demands acknowledging the counterargument. AI skills could make construction careers more attractive to younger workers who expect technology as a baseline feature of professional life. A framing crew that uses AI-driven layout tools, a plumber whose apprenticeship includes AR-guided pipe fitting, an electrician who learns to program smart panel configurations: these are real differentiators from the "same tools my grandfather used" narrative that keeps 18-year-olds from considering the trades.
NABTU's TradesFutures initiative operates over 270 Apprenticeship Readiness Programs in 34 states. If OpenAI's involvement brings resources to those programs and makes them more visible to potential recruits, that's a genuine contribution to the pipeline. Construction's image problem is real, and technology can help with perception.
But perception isn't retention. You can recruit a 20-year-old with the promise of AI-powered job sites. You keep them at 24 with predictable schedules, a clear promotion timeline, and a wage that competes with every other option their phone shows them on a bad Wednesday.
The DOL's initiative is structured as a national contract with a one-year base period and four option years. A contractor somewhere will win the bid to develop AI curricula for apprenticeship programs. Modules will be created. Pilot programs will launch. Press releases will follow. The question is whether any of it addresses the reason that half the people who start these programs never finish.
What You Can Do
If you run a residential construction company and you're considering investing in AI training for your crew, audit your attrition first. Pull your last three years of hiring and separation data. Calculate your actual attrition rate. If it's above 20%, every dollar you spend on training has a 60% chance of walking out the door within three years. Fix the leak before you fill the bucket.
If you're an apprenticeship program director, track your completion rate by trade and by year. Compare it to the DOL's national data. If your completion rate is below 50%, adding AI modules to the curriculum won't fix what's broken. Exit interviews with the people who left will.
If you're a residential homeowner wondering why your project is six weeks behind, the answer is probably not that your framing crew lacks AI skills. It's that your builder's best framer left for a competitor in March, the replacement is three months into an apprenticeship, and the apprenticeship program that trained the original framer graduated 11 people last year from a cohort of 24.
What We Don't Know
Bridgit's dataset skews large. Its 233 contractors include nearly 40% of the ENR 400, which are the biggest firms in the country. Attrition rates for small residential builders may be higher or lower. No comparable dataset exists for firms with fewer than 50 employees, which represents the vast majority of residential construction companies.
Apprenticeship completion rates from the DOL's registered system do not capture informal training, non-registered programs, or the growing number of workers who learn on the job without formal apprenticeship. The 50% completion figure applies only to the registered system.
The DOL's AI apprenticeship initiative has not yet awarded its contract. No curriculum exists to evaluate. The criticism here is of the framing and priorities of the initiative, not its execution. It's possible the selected contractor will focus heavily on retention alongside skills. The announcement didn't suggest it, but the work hasn't started.
We could not independently verify NABTU's $2.5 billion annual training investment figure. It's self-reported. Allocation across trades, regions, and program types is not published in a way that permits per-apprentice cost calculations. Our $833 per worker figure is a rough average, not a program-level cost.