Tom Kriger has a line he keeps repeating. "AI can't turn wrenches," says the director of research and education at North America's Building Trades Unions, which represents more than 3 million skilled craft workers and operates 1,900 apprenticeship training centers across the U.S. and Canada. "It's not going to lay bricks." He is right. He is also missing the point entirely, because the wrenches and the bricks are not where the problem lives anymore, and every Big Tech company writing checks for construction training knows it.
In April 2026, Microsoft announced an expanded partnership with NABTU to deliver no-cost AI literacy courses and industry-recognized credentials to construction workers nationwide. Already, 1,500 instructors had been trained across hands-on training centers, with Microsoft's LinkedIn Learning platform carrying the courses and TradesFutures, a nonprofit with apprenticeship readiness programs in 34 states, distributing them to the field.
One month earlier, OpenAI had signed its own partnership with NABTU, committing to support TradesFutures and to launch a Certifications and Jobs Platform near its Stargate data center sites in Texas, New Mexico, Ohio, and Wisconsin. Then in June, BlackRock committed $100 million over five years to train 50,000 skilled trades workers through its Future Builders initiative. That is a lot of zeroes attached to a lot of press releases in a very short window.
Follow the Money
None of these announcements are about housing. OpenAI told the Office of Science and Technology Policy that the United States will need "20 percent of its current skilled trades workforce over the next five years" to build data centers and energy infrastructure alone. Its Stargate project alone spans six massive facilities across four states, representing more than $400 billion in planned investment and 7 gigawatts of power capacity, which is the kind of infrastructure spending that demands an entirely new workforce that largely does not exist yet and cannot be trained on YouTube between concrete pours no matter how many instructors Microsoft certifies. BlackRock CEO Larry Fink has estimated that America should plan to spend $10 trillion on the data center and energy buildout required to power AI over the coming years. Future Builders money targets electricians, welders, plumbers, and HVAC technicians working on that buildout.
Residential builders short on framers do not appear.
Ask any residential GC what Microsoft is doing for their labor shortage. They will laugh.
The YouTube Curriculum
While the partnership announcements pile up, here is what construction workers are actually doing to learn AI. A DEWALT study published in April 2026 surveyed the industry and found 90 percent of construction professionals believe AI will become indispensable within five years. Eight percent use it today. Not eight percent in a trial program or eight percent with formal training, but eight percent doing anything at all with AI in their daily work.
Eighty-seven percent said AI education should be embedded in trade schools and technical programs, which sounds like the kind of consensus that should move policy, but nobody is waiting. They learn from YouTube videos, online forums, and whatever free tools they stumble across between jobs, which is to say they are building their own curriculum out of whatever the algorithm recommends after a ten-second search on a cracked phone screen during lunch. Fifty-nine percent specifically asked for hands-on, task-based training rather than lectures or credentials, the kind of learning you cannot deliver through a LinkedIn Learning module no matter how many instructors you certify to distribute it.
A ServiceTitan report published in 2026 found 74 percent of residential contractors view AI as an efficiency engine, but only 25 percent are using it at all, and the early adopters report measurable results that should embarrass the holdouts: 48 percent saw increased productivity, 45 percent reported time savings.
The Per-Worker Math Nobody Publishes
NABTU invests roughly $2 billion per year in its training programs across 1,900 centers, which works out to approximately $667 per worker per year for 3 million members. That money goes toward apprenticeship instruction in welding, electrical, pipefitting, and every other physical trade that keeps buildings standing, and it is funded jointly by unions and their contractor partners without a dollar of taxpayer money.
BlackRock's $100 million over five years sounds enormous in a press release, but divide by five and you get $20 million per year, and divide that by NABTU's 3 million members and you get $6.67 per worker annually, which is one percent of the existing per-worker training investment. Its target is 50,000 workers total, which is 1.7 percent of NABTU's membership and 0.44 percent of the 11.4 million construction workers in the United States. No cash there. Microsoft and OpenAI's contributions are primarily in-kind: platform access, course development, and credential design rather than direct funding.
| Investment | Annual Amount | Per Worker (3M NABTU) | Per Worker (11.4M total) |
|---|---|---|---|
| NABTU existing training | $2 billion | $667 | $175 |
| BlackRock Future Builders | $20 million | $6.67 | $1.75 |
The money is real.
It is also a rounding error.
What the Apprentices Already Know
The most interesting finding from the NABTU-Microsoft program is not the curriculum. It is the focus groups. When Microsoft and NABTU sat down with apprentices to assess their needs, they found something the training centers had not anticipated: younger workers were already using AI tools on job sites without any formal instruction, interpreting electrical codes through ChatGPT, checking OSHA regulation updates on their phones between tasks, troubleshooting equipment problems with AI assistants during breaks. Nobody told them to do it, nobody trained them, and they just did it, because the tools were free and the alternative was paging through a thousand-page codebook on a ladder.
"That's the whole goal, to make our instructors more efficient so they can spend more time with apprentices," Kriger told Construction Dive. Instructors use AI to generate lesson plans, quizzes, and training materials, which is useful, but the apprentices are using AI to solve actual construction problems on actual job sites, and nobody certified them to do that, and in many cases the instructors are learning from the students. Adoption moved faster than any training program designed to teach it.
The Uncomfortable Question for Homebuilders
If you are building a home in 2026, your general contractor is statistically unlikely to use AI in any meaningful way, and three out of four don't. At 25 percent adoption among residential contractors, three out of four builders are estimating by hand, scheduling on instinct, and managing code compliance the way they have for decades. That is fine until it is not, and it stops being fine when the builder across town figures out that AI-assisted scheduling compresses timelines by 15 to 20 percent and AI takeoff tools cut estimating time in half.
None of this money is aimed at residential construction, not directly, but it is flowing toward data centers, power plants, and the energy grid, because that is where the demand is explosive and the labor constraint is existential. But workers trained in AI literacy for data center projects do not unlearn those skills when they move to a residential job, and they bring habits that the residential industry never taught them. Spillover is real, and it may be the most honest path to AI adoption in homebuilding: not from top-down industry programs, but from workers who picked it up building something else.
Limitations of This Analysis
The per-worker calculations divide total announced investment by total workforce size, which overstates the dilution because no program claims to reach every worker, and understates the concentrated impact on the workers it does reach. At $2,000 per worker within BlackRock's 50,000-worker target, it is a meaningful investment that compares favorably with community college tuition for a technical certificate, though the program's curriculum and completion rates are not yet public. DEWALT's survey methodology, sample size, and respondent demographics were not disclosed in their published summary, and the 8 percent adoption figure may not represent a statistically rigorous cross-section of the industry. Whether data-center-trained workers carry AI skills into residential work is logical but unproven, and it is possible that the AI tools used in large-scale infrastructure construction have limited applicability to the fragmented, relationship-driven workflow of residential building.