On March 11, Sam Altman stood next to Sean McGarvey, president of North America's Building Trades Unions, and announced a partnership to "help support and expand training pathways into the skilled construction trades." Six weeks later, Microsoft signed its own expanded deal with NABTU, rolling out free AI literacy courses through LinkedIn Learning and touting an initiative that had already reached 1,500 instructors. On April 29, the Department of Labor launched an AI in Registered Apprenticeship Innovation Portal. Three announcements in seven weeks, and the construction industry's AI training revolution had supposedly arrived.
It hadn't.
What arrived was not a revolution but a press strategy, one with logos and handshakes and free LinkedIn Learning accounts and almost no construction-specific content underneath any of it.
What the Partnerships Actually Contain
Read the NABTU-OpenAI announcement carefully and you will find one phrase buried in the third paragraph that explains the entire deal: "project entitlement." In construction, that term means permits. Specifically, it means getting local governments, community groups, and labor organizations to approve your project before a single shovel hits dirt, to sign off on the zoning variances and environmental impact reports and community benefit agreements that stand between a company and a completed building. OpenAI is not building an AI curriculum for electricians. OpenAI is building data centers, and it needs union labor and union political support to get the permits signed.
Microsoft's deal with NABTU is more concrete, which makes it easier to evaluate. It rolls out in phases: data security basics first, then AI literacy, then what they call "practical applications." Those practical applications, as described by Tom Kriger, NABTU's director of research and education in Construction Dive, focus on helping instructors generate lesson plans, quizzes, and training materials more efficiently. "That's the whole goal," Kriger said, "to make our instructors more efficient so they can spend more time with apprentices."
That is a reasonable use of AI. Genuinely. It is also instructor productivity software, not workforce transformation. A journeyman ironworker is not learning to use AI for structural load calculations. A plumber is not getting trained on computer vision for pipe inspection. Instructors are getting a tool to write quizzes faster.
The Workers Who Are Actually Using AI Learned on Their Own
The most revealing detail in the Construction Dive report is a throwaway line about focus groups. When NABTU conducted sessions with apprentices, they discovered that many younger workers were already using AI tools on job sites, unprompted, without any training program. They were pasting electrical codes into ChatGPT to get plain-language interpretations. They were asking AI to reconcile changes between OSHA standard revisions that would have taken hours of manual cross-referencing.
Nobody taught them to do this, no corporate partnership funded it, and no LinkedIn Learning module covered it. They opened a phone app and typed a question. That is it. That is the entire workflow, and the entire formal training apparatus, with its press releases and its credential programs and its phased curriculum rollouts, is trying to systematize something that is already happening organically, at a pace and specificity that no top-down program can match, because the worker who needs to know whether a 200-amp panel feeds a subpanel in compliance with 2023 NEC 408.36 does not need an AI literacy credential to type that question into a chat window.
"AI can't turn wrenches and it's not going to lay bricks," Kriger told Construction Dive. He is right, but the workers who are using AI to look up codes, troubleshoot installation sequences, and check safety compliance updates did not need a partnership announcement to figure that out.
The Numbers That Don't Add Up
The Associated Builders and Contractors published its workforce demand model for 2026. Construction needs 349,000 new workers this year. By 2027, that number rises to 456,000. Construction added only 15,000 jobs in all of 2025, its weakest year since 2011.
Against that demand, consider the training pipeline. TradesFutures, the pre-apprenticeship nonprofit both OpenAI and Microsoft are supporting, enrolls 7,700 people per year across 34 states. The U.S. Department of Labor reported in 2021 that overall apprenticeship completion rates are below 35 percent. Apply that rate to TradesFutures: 7,700 enrollees times 35 percent completion yields roughly 2,700 workers per year who actually finish the program and enter the trades.
| Pipeline Component | Annual Output | Share of 349K Gap |
|---|---|---|
| TradesFutures completers (est.) | ~2,700 | 0.77% |
| NABTU instructors trained (total to date) | 1,500 | N/A (instructors, not field workers) |
| All government-registered apprenticeship completers (2023) | 40,000–45,000 | 11.5–12.9% |
| Workers needed (2026) | 349,000 | 100% |
Even if every single government-registered apprenticeship completer in the country entered construction, the pipeline would cover 13 percent of the annual demand. Adding an AI literacy module to that pipeline does not change the throughput; it changes the press release.
Follow the Money to the Data Center
OpenAI's Stargate project, a joint venture with SoftBank, involves more than $100 billion in planned data center infrastructure across the United States. Microsoft committed over $80 billion in AI infrastructure spending for fiscal year 2025 alone. These projects require union labor, and union labor requires NABTU sign-off, and NABTU sign-off requires something in return. What NABTU gets in return is training partnerships that elevate the trades' public profile, funding for TradesFutures, and the political leverage that comes from being the official workforce development partner of the two most valuable companies in the world.
This is how large-scale construction has always worked. Nobody builds a $10 billion data center campus in a congressional district without making sure the local building trades council is on board. The training partnerships are a component of community benefit agreements, which are themselves a component of project entitlement. That pipeline runs from Altman's handshake to McGarvey's endorsement to a county supervisor's permit approval. Follow that chain.
None of this is corrupt, but all of it is transactional. And the transaction is not really about teaching construction workers to use AI in their daily work. It is about building data centers with union labor while generating enough public goodwill that nobody objects to the zoning variance.
What Would Real AI Training for Construction Look Like?
If you wanted to build an AI training program that actually changed how residential construction workers do their jobs, you would start with the problems they have today, not the problems tech companies want to solve. An electrician does not need an AI literacy credential. An electrician needs a tool that cross-references the 2023 NEC amendments with their local jurisdiction's adopted code year and tells them whether a particular panel configuration is compliant before the inspector shows up, because right now that process involves a physical code book, a phone call to the building department, and sometimes a second trip to the site.
A framing carpenter needs an app that photographs a delivered lumber stack and flags warped or undersized boards before they get installed, because a 2x4 that is actually 1.48 by 3.42 inches instead of the nominal 1.5 by 3.5 will pass a visual check but fail a load calculation, and catching that at delivery saves $2,000 in tearout. A superintendent needs a scheduling tool that ingests weather data, subcontractor availability, and inspection queues and produces a critical path update every morning, because right now that superintendent is spending 90 minutes before sunrise on a spreadsheet that is already wrong by the time the first truck arrives.
These tools exist in various stages of development. None of them are part of the NABTU-Microsoft curriculum. What they built is AI literacy. What the job site needs is AI plumbing. Actual plumbing. Tools that solve the problems construction workers already have, not courses explaining what artificial intelligence is to people who have been using it on their phones for months.
The Strongest Case for the Partnerships
In their defense: something is better than nothing, and the baseline is genuinely nothing. Most construction workers have received zero training on how AI will affect their work, their job security, or their industry. A PYMNTS Intelligence study published in April found that nearly half of all salaried or higher-paying workers received no on-the-job AI training in the past 12 months. Among hourly construction workers, the number is almost certainly worse. Microsoft's free LinkedIn Learning courses eliminate a cost barrier. The AI literacy credential, while not a trade-specific skill, signals to employers that a worker has baseline digital fluency, which increasingly matters for technology-adjacent roles like estimating, project coordination, and safety management.
And the political argument has genuine weight: NABTU's $2.5 billion annual training investment, funded entirely through collectively bargained private-sector contributions, is the largest workforce development pipeline in any domestic industry. If these partnerships channel even a fraction of Big Tech's infrastructure spending into expanding that pipeline, the 7,700 annual TradesFutures enrollees could become 20,000 or 50,000. Scale changes everything in workforce development, and right now the pipeline is subscale by an order of magnitude. That gap is real.
If You Are Building or Hiring a Builder
Do not evaluate a contractor based on whether their crew has AI training credentials. Evaluate them based on whether they are using any digital tools for code compliance, scheduling, or quality control. Contractors who adopted Procore, Buildertrend, or CoConstruct five years ago are the ones whose crews are now informally adopting AI. What matters is not the tool but the culture of tool adoption, and that culture is not created by a LinkedIn Learning module. It is created by a superintendent who hands an apprentice a tablet and says: check this.
If you are a builder running a crew of 10 to 20 workers and you want to introduce AI tools, skip the literacy courses. Buy your office manager a ChatGPT subscription and have them use it for takeoff verification, change order language review, and subcontractor communication drafting. That costs $20 a month and will save your office 5 to 10 hours a week within 30 days, no partnership required.
Limitations
This analysis relies on publicly available press releases, news coverage, and secondary sources rather than direct access to the NABTU-Microsoft curriculum materials, which are distributed through JATC training centers and not publicly posted. The 35 percent apprenticeship completion rate is a national average from DOL 2021 data, and completion rates vary substantially by trade, union, region, and program structure. The characterization of these partnerships as primarily motivated by data center permitting is an inference from publicly stated priorities and spending patterns, not from internal communications or stated intent by any party. The TradesFutures pipeline math uses rough estimates; actual output could be higher if completion rates for that specific program exceed the national average, which is plausible given NABTU's comparatively well-resourced training infrastructure. Microsoft's curriculum may evolve to include trade-specific AI tools that are not yet reflected in public reporting.