Meta, Google and BlackRock have committed about $265 million to train skilled trades workers. The same companies will spend more than $335 billion on data centers this year.
Alphabet and Meta will spend somewhere north of $335 billion on capital projects in 2026. Their combined announced commitments to training the electricians, welders and fiber technicians who physically build those projects come to roughly $265 million, which works out to about seven hours of that capital spending. The gap is the story. Labor is now the input the AI industry cannot buy its way past, and the money it is putting into fixing that says the companies know it.
Tyler Shelton is 29 and spends his days climbing down manholes in Detroit to fix cables. His employer has been sending crews to a data center that broke ground an hour away, and he told The New York Times that everyone wants that money, even as his own company loses manpower to the site. The project he might join is OpenAI’s campus in Saline Township, which the state of Michigan calls the largest single investment in its history. Hundreds of electricians, ten-hour days, seven days a week.
Shelton wants to buy a house in the next year. He also wants a life outside work. Both things are true, and the tension between them is what the AI industry is now trying to price.
What Happened
Lydia DePillis reported for The New York Times on July 29 that AI companies have moved from buying construction labor to manufacturing it. Three commitments anchor the story.
Meta launched America’s Workforce Academy on June 8 with $115 million for its first year, targeting about 5,000 participants. Trainees get a four-week course with travel, lodging and a stipend covered, an industry credential from the National Center for Construction Education and Research, and a guaranteed job offer from a contractor building Meta data centers. The program runs with Associated Builders and Contractors, a nonunion trade association with 67 chapters, and pilots in Baton Rouge, Columbus, Indianapolis and Houston.
Google took the opposite route. Its $50 million program flows through the International Brotherhood of Electrical Workers and its contractor partners, funding 14 unions and four trade associations across 20 states. The goal is to lift annual apprenticeship enrollment from 19,500 to 30,000 for three years in locations Google picks. Maggie Johnson of Google.org framed it as getting people into the trades wherever they want to work.
BlackRock is funding training pipelines around its Texas data centers as part of a $100 million skilled-trades effort. The asset manager already sits at the center of this buildout in a different capacity, as the 80% owner of Meta’s $12 billion El Paso data center.
Tina Williams, who manages the grants for the IBEW alliance, described the balancing act plainly: bring in enough workers for the data centers while keeping the core work in each local staffed.
The Backstory
Brad Smith of Microsoft has said the electrician shortage is the single biggest obstacle to expanding data centers in the United States. Microsoft has been flying electricians in from 75 miles out to keep sites moving. It started training data center technicians back in 2018 and now runs 39 locations that have put 15,000 people through.
The arithmetic behind the shortage is not subtle. The Bureau of Labor Statistics counted 818,700 working electricians in 2024 and projects about 81,000 openings a year through 2034. Associated Builders and Contractors puts the construction industry’s 2026 need at roughly 349,000 additional workers on top of normal hiring. Close to a third of union electricians are near retirement.
Then the AI capex cycle landed on top of that. Alphabet raised its 2026 capital expenditure guidance to $195 billion to $205 billion, spent $44.9 billion in the second quarter alone, and posted negative free cash flow of $5.9 billion. We covered that print and why the $514 billion cloud backlog did not save the stock. Meta raised its own 2026 range in April to $125 billion to $145 billion, citing higher component prices and additional data center costs, and reports second-quarter results today.
Turner & Townsend found data center construction labor costs in primary North American markets rising 8% to 12% a year, driven by trades scarcity. Indeed’s analysis found hourly installation and maintenance roles at data centers paying 42% more than comparable jobs elsewhere. Marty Schager at the staffing firm Aerotek called it a delicate labor tension, with a passive job-seeker community waiting to capture a once-in-a-generation opportunity.
The Plan
Two companies looked at the same bottleneck and gave opposite make-versus-buy answers.
Meta is building captive supply. Four weeks, a credential, a stipend and a job on a Meta site. Joel Thames of ABC described the goal without much varnish: get as many of these people onto Meta sites as possible and keep them in the family. If someone leaves for another sector, the program treats that as acceptable rather than desirable. Meta’s earlier fiber technician program, Level-Up, pulled 35,000 applications in its first week, so demand for the on-ramp is not the constraint.
Google is subsidizing the commons. Its money goes into an apprenticeship system that already exists, produces journeymen over four years, and sends those workers wherever the work is. Sean McGarvey, president of North America’s Building Trades Unions, represents three million workers whose affiliates spend about $2.5 billion a year on apprenticeships and continuing education, funded by employers and members. He called Meta’s four-week program a brilliant public relations move and said comparing it to a four-year apprenticeship is apples and oranges.
McGarvey also made the claim that matters most. His group can train as many people as the AI companies need, provided those companies commit early to hiring union labor. OpenAI made that commitment in March.
The Business Model Angle
Every other input in this buildout responds to money. Chips: order more, pay Nvidia’s margin. Land: buy it. Power: sign a purchase agreement, fund a turbine, revive a reactor. Financing: structure a joint venture and sell $27 billion of private credit against it. Capital converts into all of those on a timeline the buyer controls.
A journeyman electrician takes four years. No check compresses that, which makes labor the first input in this cycle where the hyperscalers cannot outspend the problem. So they are running three plays at once, and only one of them creates new workers.
The first play is the wage premium. Paying 42% above market moves electricians from housing, hospitals and factories onto data center sites. That is reallocation, not creation. Mario Iacobacci of Oxford Economics put it in terms any operator recognizes: resources are limited, so building one thing drags from another. The premium is also self-inflicted cost. Labor inflation of 8% to 12% a year on data center sites feeds directly into the capex lines that investors have started punishing, which is part of the story behind Meta’s $10 billion guidance raise in April.
The second play is the four-week shortcut, which produces workers who can do specific tasks on a specific kind of site. Useful for schedule risk. Not a substitute for the pipeline.
The third play is the one the training budgets actually reveal. Set Meta’s $115 million against its own capex and you get about eight basis points. Set it against the $10 billion the company added to its 2026 capex range in a single April announcement and the workforce academy is 1.2% of the revision. Nobody solves a 349,000-worker shortage with a rounding error. What $115 million does buy is local endorsement, and Anthony Abrantes of the Eastern Atlantic States carpenters explained the mechanism: unions endorse projects they are set to work on and sit out the debates when they are not. He said the trades are doing the developers’ advocacy and business development for them.
Offshore wind is the cautionary case both sides keep citing. That industry lost its permits to organized local opposition, and Abrantes said it did a bad job of educating communities. Data center developers watched it happen. Meta has been running America’s Workforce Academy ads across Facebook and Instagram and working with the National Urban League and the U.S. Hispanic Chamber of Commerce to spread the word. Ramiro Cavazos of the chamber wants more Hispanic contractors hired and thinks better local relationships could smooth Meta’s path in skeptical places.
Read as a permitting expense rather than an HR expense, $115 million against a $135 billion capex program stops looking small. It looks like cheap insurance on the one risk that has actually killed an American infrastructure boom in the last decade.
The Risk
The credential travels. The job guarantee does not. Meta’s graduates get an NCCER card that works across employers, and a placement that exists as long as Meta keeps pouring concrete. Joe Ottenbacher, who left early childhood education for the trade, asked the question that hangs over the whole thing: when the builds finish, where do the workers go, and what happens to wages when they all arrive at once?
Jeff Strohl at Georgetown’s Center on Education and the Workforce gave the optimistic version, where the data center boom cools right as a housing boom starts and absorbs everyone. He also said that is probably not likely.
Then there is the possibility that the shortage is not a training problem at all. McGarvey says the union system can scale to whatever the AI companies need, as long as they commit to union labor early enough for the locals to plan. If he is right, the binding constraint is contractual rather than physical, and the four-week academies are solving a problem that a signed project labor agreement would solve better. That is a self-interested claim from a man who runs the alternative. It is also the only claim in the story backed by a $2.5 billion annual training budget that already exists.
Public patience is the third risk. Protesters met President Trump outside Detroit this week over data centers. Starr Sciortino, 22, in her second year of the IBEW apprenticeship, compared working on a data center to helping build a weapons factory and said she does not believe any amount of money justifies sacrificing community resources. She is exactly the worker these programs are built to recruit.
Quick Questions
How much are AI companies spending on skilled trades training? Meta committed $115 million for the first year of America’s Workforce Academy, Google $50 million over three years through the IBEW and its contractor partners, and BlackRock $100 million toward skilled trades including pipelines for its Texas data centers. The total sits around $265 million against combined 2026 capex above $335 billion for Alphabet and Meta alone.
Why can’t AI companies just pay more for electricians? They are, at about 42% above comparable jobs according to Indeed. Higher pay moves existing electricians between projects without adding any to the national pool, and it raises construction costs on the sites doing the bidding.
How long does it take to train an electrician? A union apprenticeship runs four years and produces a journeyman. Meta’s program runs four weeks and produces a credentialed entry-level construction worker. Sean McGarvey of North America’s Building Trades Unions calls the comparison apples and oranges.
What happens to these workers when the data center boom ends? Nobody has an answer. Fully trained electricians can move to nuclear plants, apartment buildings or pharmaceutical factories, but no other sector is building at this scale. Workers interviewed by the Times raised the risk that a sudden influx of available electricians pushes wages down for the whole trade.
Is the data center labor shortage real or a negotiating position? Both. Contractors, staffing firms and Microsoft’s own president describe a hard constraint. The building trades argue they can meet the demand if developers commit to union labor early, which turns the shortage into a coordination problem rather than a supply problem.
The Business Model Analyst Take
Watch where a company puts its money when it hits a wall it cannot buy through. Meta’s answer is a $115 million training academy and an ad campaign. Its actual answer is $135 billion of capex and a 42% wage premium that pulls crews off other people’s job sites.
The training programs are worth having. They are also priced like marketing, structured like recruiting, and timed like a permitting strategy. The industry has spent three years converting payroll into compute, and Meta has been explicit about that trade. The irony arriving now is that compute turns out to need payroll, in a trade nobody can automate and nobody can accelerate.
For anyone underwriting one of these campuses, the number to track is not the capex figure or the training headline. It is the wage line on the contractor’s bid, and whether the crews on site came from a pipeline these companies built or from the hospital project down the road. One of those is expansion. The other is a transfer, and transfers generate the local opposition that has already killed one American energy buildout this decade. Meta is spending $115 million to make sure it does not kill this one.
