A few days ago, Musk posted a seemingly ordinary recruitment message on X, saying he was looking for “skilled technicians” to build “the most powerful AI supercomputing cluster on Earth and beyond.” Building data centers in space is nothing new—Musk has mentioned it many times—but the phrase “skilled technicians” was striking. Clicking into the job posting and reading through the discussions in U.S. tech circles and labor unions, the result was startling: Americans are now so desperate to build AI that they are running out of electricians.
1. American electricians’ value is skyrocketing
The first sign came from data released by the International Brotherhood of Electrical Workers: at some data centers, the number of electricians required is two to four times the total registered members of the local union—even if every single worker in the area were split into three people, it still might not be enough. When the top U.S. tech magazine WIRED interviewed contractors, it also found that a technician had just finished training at one data center, and the construction team at another nearby site was already poaching him with signing bonuses, night‑shift premiums, and cross‑state travel allowances. After the “post‑war golden age,” are American blue‑collar workers finally going to be great again? How夸张 are their salaries?
According to the electrical workers’ union, construction workers at data centers earn an average annual income of $81,800—fully 32% higher than on ordinary construction projects. But that $81,000 is just the average. If they work as hard and put in as much overtime as Chinese workers, at many data centers American electricians can earn as much as $240,000 to $280,000 a year. The United Association of Plumbers and Pipefitters came right out and said: Silicon Valley algorithm engineers are nothing, top university scientists are nothing—without electricians, none of it works. The technicians sweating on the construction site, they declared, “are the real battlefield for AI talent.” The founder of LinkedIn gave this “rise of electricians” an even more provocative name: “the gold rush of AI infrastructure.”
The reasons for the shortage are not hard to understand. In the past, internet data centers had modest computing demands, and ordinary server racks could get by with a simple plug‑in. Today’s AI data centers, however, are like industrial monsters that devour electricity and water. They contain not only GPUs, but also large substations, medium‑ and high‑voltage distribution, UPS systems, backup generators, busbars, control systems, and a host of other power facilities. The larger the model, the more chips it needs; electricity consumption soars, heat generation is extreme, and liquid‑cooling systems far more complex than before must be installed. So AI is not just competing for electricians—it is also competing for HVAC workers, pipefitters, and large numbers of welders, sheet‑metal workers, fiber‑optic installers. In short, every skilled trade related to electricity, HVAC, welding, piping, or communications equipment is now among the scarcest resources in this “AI infrastructure gold rush.”
2. Big money from tech giants, fast‑track to employment
As everyone knows, the United States already suffers from a chronic shortage of workers who can actually do the job. So things started to turn surreal. Google was the first to panic, announcing that if there were not enough electricians, it would pay to train them. It made a high‑profile pledge to support ETA, the largest electrical worker training system in the U.S., by providing training for 100,000 existing electrical workers and cultivating 30,000 new apprentices, with the goal of expanding the pool of electricians capable of building AI data centers by 70% within five years. Most dramatically, Google even inserted AI courses into electrician training—first teach electricians to use AI, then let them build AI data centers, creating a closed loop.
Microsoft was not idle either. Besides funding electrician training, it set up training labs with decommissioned data‑center equipment to help veteran electricians make the transition and get used to data‑center work as soon as possible. OpenAI went even more direct. It has been pushing its “Stargate Project,” planning to build multiple data centers across the country, and was so short‑handed that early this year it announced it would set up training classes tailored to each data‑center location, not only cultivating talent but also offering completion certificates and job referrals. Meta, as usual, arrived late but opened its wallet generously, as if unsure how to spend its money. To train electricians, Meta clearly stated it would invest $115 million in the first year; trainees would not only attend free of charge but also receive stipends during the training period. The course lasts only five weeks, and upon completion, graduates not only get a certificate but also a guaranteed job—not just a referral, but an actual position. At this rate, if American electricians showed up at a People’s Park matchmaking corner, even the aunties would be fighting over them.
While the big companies were scrambling for solutions, in January this year Jensen Huang, the king of AI hardware, declared at the Davos Forum that AI is “the largest infrastructure build‑out in human history,” saying that chip factories, computer factories, and AI factories would unleash a huge number of jobs for electricians, plumbers, steelworkers, and construction workers. One wonders whether he realized at the time that the problem was not just the number of jobs—there simply are not that many electricians in the whole United States, and you cannot just pull them from hospitals and schools.
For a while, only the tech companies were anxious. A few months later, financial capital could no longer sit still. BlackRock, the world’s largest asset manager, announced it would spend $100 million over five years to help 50,000 people complete training in electrical, HVAC, plumbing, and other skills. The logic was absurd yet simple: pouring money into data centers is useless; Wall Street elites can conjure up hundreds of billions of dollars through countless financial instruments, but they cannot conjure up a single certified electrician.
3. Is college less worthwhile than trade school?
Given this situation, young Americans are not stupid. They have completely seen through the new employment logic of the era: taking out student loans to study programming, graduating, and then competing with AI for entry‑level positions—that is probably not as good as taking a five‑week trade course, putting on a hard hat, and wiring up AI. There is indeed a lot of such discussion on X.
It sounds great, and China is much the same. As early as 2023, the China Academy of Information and Communications Technology pointed out that AI data centers are facing the challenge of “large scale, high growth, and urgent delivery,” with a shortage of mature operations and maintenance talent, especially in electrical, HVAC, automation, and security. But hold on—especially those high‑school and college students already searching for nearby trade schools, and coders planning to switch to electrical work. What AI data centers lack is not the electrician who does home renovations; they need real electrical and HVAC engineers who can handle high‑voltage distribution, redundant power supplies, liquid cooling, automation, and fault‑free maintenance. Baidu’s public job postings make it clear: data‑center electrical and HVAC engineers require a bachelor’s degree or above, plus more than five years of relevant experience. So the university and conventional career path probably still have to be followed.
Nevertheless, a clear trend has emerged: among Tsinghua University graduates, the number going into manufacturing and energy industries increased by 19.1% year‑on‑year in 2025, and has been growing for several consecutive years. Chinese students’ career directions are indeed shifting from pure internet and finance toward hardware, energy, power, and advanced manufacturing. As for coders already in the workforce, with the development of world models and embodied intelligence over the past two years, since last year they have gradually been trying to connect their code to the physical world, moving into industrial software, energy management, robotics, data‑center automation, DCIM, BMS, and on‑site technical services. This is not coders turning into electricians; it is shifting from writing code for apps to writing code for power grids, equipment rooms, and machines. So AI has not made knowledge useless; it is simply repricing knowledge.
At this point, you are probably curious how American electricians themselves view this “rise.” When WIRED visited U.S. electrician communities, it found that some workers, even offered sky‑high pay, explicitly refused data‑center projects. Their reason: they felt they were building a machine that might make white‑collar workers jobless, drive up local electricity rates, and add to the environmental burden. And so the most surreal scene of the AI era appears: the very people least likely to be replaced by AI are now hesitating about whether to plug AI in with their own hands.
