Meta is poaching electricians with seven-figure annual salaries, and is so desperate for skilled workers that it has set up its own vocational school.

Following Chengxiang's analysis, the new bottleneck in the AI competition has arrived at construction sites.

On an early morning in the fields of Saline, Michigan, hundreds of electricians and construction workers show up on time, then work consecutively for ten hours, repeating this routine seven days a week.

What they are building is the latest campus of OpenAI's "Stargate" project, a $16 billion ultra-large-scale data center.

According to McKinsey's workforce forecast, for AI infrastructure expansion alone, the United States will need to additionally train 130,000 electricians, 240,000 construction workers and 150,000 construction supervisors between 2023 and 2030.

The U.S. Bureau of Labor Statistics estimates that between 2024 and 2034, there will be 80,000 unfilled electrician positions every year.

Even compared with traditional housing, healthcare and energy sectors, AI companies are very willing to pay high premiums.

For example, the salary for a short-term maintenance position in an AI data center can be 42% higher than that in traditional sectors, and an excellent electrician can earn an annual salary as high as $240,000 to $280,000.

However, this amount of money is hardly worth mentioning compared with the online project losses of large AI companies.

After all, the construction of electrical systems accounts for 45% to 70% of the total cost of the entire data center. For a single 60-megawatt project, delivery delays caused by labor shortages will directly bring $14.2 million in revenue losses every month.

To this end, Microsoft President Brad Smith publicly stated: "The electrician shortage has become the number one barrier for us to expand data centers in the United States."

Why do AI companies need so many skilled workers?

High-intensity working conditions are inevitable for the construction of AI data centers.

AI companies do not "need more construction workers", but need more "technical experts who are highly skilled and can withstand high-intensity work".

The construction difficulty of AI data centers far exceeds imagination. There is almost no comparable precedent for their construction complexity.

This complexity is mainly reflected in three aspects.

The top priority is the huge power consumption required by AI data centers. After all, the power consumption of a single GPU rack has reached 120 to 140 kilowatts, ten times that of the standard server racks ten years ago.

A typical ultra-large-scale AI data center may deploy tens of thousands or even 100,000 GPUs, and the power consumption of the entire facility often reaches hundreds of megawatts, which is equivalent to supplying power to hundreds of thousands of households at the same time.

Sam Altman once posted that the picture shows the Stargate Base 1

The next is the power distribution system with extremely complex structure.

Such huge power consumption means that the entire power distribution system must be designed from scratch, including switchgear, transformers, uninterruptible power supply systems, bus ducts, cooling loops and so on. Subsequent installation and commissioning must be completed by certified professionals, and no software can replace this process.

In addition, high power consumption also brings heat dissipation problems: the heat density of AI facilities has exceeded the limit of air cooling, and must turn to direct liquid cooling and immersion cooling, whose design and installation also rely on certified plumbers and HVAC engineers.

Therefore, not only for electricians, the number of job vacancies for HVAC engineers in the United States also increased by 78% between 2022 and 2026.

Tech giants are poaching workers again, and electrician training starts from teenagers

Young and skilled electricians have become the targets that tech giants are poaching from each other.

These electricians have been poached 3 times in just 18 months, just like "drafting in the Major League".

In 2026, Alphabet and Meta announced that their total investment in skilled worker training will reach $265 million (even though their total capital expenditure has reached $335 billion).

Facing the shortage of skilled workers, it is better to train workers by themselves than to recruit them externally. These AI companies known for their speed cannot wait for death, and they have already taken action.

For example, Meta allocated $115 million to launch a long-term construction worker training school, with the first batch of 5,000 trainees. During the period, tuition fees, air tickets and accommodation are all free, and there are additional living allowances. After a four-week training period, the trainees will start working immediately.

Meta has flooded Facebook and Instagram with ads for this program, eagerly inviting young people to enroll.

Meta technical school advertisement

On the other side, in March this year, OpenAI reached a cooperation with North America's building trade unions, promising in advance to hire union workers to obtain more skilled construction workers with longer training experience.

This "gold rush" for skilled workers has spread to high schools. These AI companies are eager to find high school students who are about to graduate, promising them a great opportunity to make big money.

Kid, you have to be an electrician. This is a really well-paying job. Come on over quickly!

The results are undoubtedly very remarkable.

Data shows that between 2017 and 2025, the enrollment rate of Generation Z in technical schools increased by 1421% (this high growth may come from a relatively low statistical base). A 2025 survey found that 42% of Generation Z are willing to consider skipping college and entering skilled worker training directly. According to data from this year, 60% of Generation Z have already planned to engage in technical work.

How much more electricity will AI training consume in the future?

The electricity issue has always been hanging over the desks of executives of AI technology companies.

Nowadays, AI data centers are devouring electricity at an astonishing speed.

Data from June 2026 shows that global data center power consumption will reach 565 terawatt-hours this year, a year-on-year increase of 26.4%. This growth rate almost all comes from AI servers, which account for only a small part of the installed capacity but consume 31% of the total electricity, and this number is still expanding at an annual rate of 84.2%.

Data from the International Energy Agency (IEA) is more intuitive: in 2025, the power consumption growth rate of AI data centers is 16 times the growth rate of global overall power consumption. Some institutions predict that by 2028, data center power consumption will account for 6.7% to 12% of the total power consumption in the United States.

IEA's forecast for global data center power demand

These electricity bills have been averaged into the bills of American residents. Analysis found that electricity prices in areas with dense data centers have increased by up to 267% compared with five years ago. The independent market monitor of PJM, the largest power grid in the United States, calculated that in the 12 months starting from June 2024, data centers added more than $9.3 billion in additional electricity costs to consumers between Illinois and Washington, D.C.

Then who will build the infrastructure that carries all this electricity? Are there enough electricians?

In addition, data center construction positions are project-based. This means that the number of workers required for construction projects and operation projects is not proportional.

The Stargate data center in Texas has a power consumption of 1000 megawatts, which is equivalent to the output power of a nuclear reactor. It requires 6,400 workers working simultaneously during the construction period, but only 100 to 1,000 permanent workers are needed for daily operation.

The image is generated by AI

Therefore, after that, hundreds of thousands of trained skilled workers will pour into other fields one after another, thus pushing down the general wage level of the industry.

The best case scenario is: when these data centers are completed and put into operation, it just catches up with the real estate boom.

But the question is, is that possible?

References:

[1]https://www.nytimes.com/2026/07/29/business/economy/data-center-electricians-training.html

[2]https://qz.com/ai-data-center-electricians-carpenters-meta-google-1851690612

[3]https://www.techtimes.com/articles/310086/20260622/ai-boom-needs-130-000-more-electricians.htm

[4]https://www.irecruit.co/insights/data-center-construction-labor-market-report

[5]https://qz.com/ai-electricians-data-center-boom-gen-z-trades-1851685432

[6]https://therinvio.com/electrician-shortage-data-center-boom/

[7]https://axis-intelligence.com/ai-data-center-energy-consumption-statistics/

This article is from the WeChat official account "QbitAI" (ID: QbitAI), written by Cheng Qian, authorized for release by 36Kr.