The Boom Is Now Paying Electricians Double

In Abilene, Texas, the electricians wiring OpenAI's Stargate data center are sometimes being paid double the going rate for their trade ^[1]. A single pay stub is not a labor report, but this one reads like a clue that something in the economy's pricing has quietly flipped. Look at the other side of the same boom and the numbers run the opposite way. Apollo Global Management finds workers in AI-exposed occupations saw real-wage growth 6.7 percentage points slower than less-exposed workers after 2023 ^[2]. Software postings have grayed: nearly 70 percent now ask for senior hires, up from 55 percent in early 2019, the highest concentration of any of the more than forty sectors Indeed tracks ^[3]. The junior rung that once priced a new graduate's first code has largely vanished, absorbed by models that write the routine work themselves. The reason both sides are one story is something the industry states out loud. Jensen Huang, whose company sells the chips beneath all of it, now describes AI as a five-layer cake: energy at the bottom, chips next, then the physical plant, then the models, then the applications. Put the models fourth and the point is hard to miss.

Energy is the first principle of AI infrastructure and the binding constraint on how much intelligence the system can produce. — Jensen Huang

That is not a line thrown off on a stage. General Electric's gas turbines are sold out through 2029 ^[4], and Nebius, a firm that rents frontier compute, says demand still exceeds what it can supply ^[5]. The scarcest thing in this industry right now is not a smarter model. It is megawatts, concrete, and hands. The hands are the part the labor market is pricing with unusual force. American chipmakers face a projected 157,000-worker shortage by 2030 and are importing talent from Asia because the domestic pipeline is thin; Samsung says it simply does not see enough technical people coming through ^[6]. Even Taiwan, which makes most of the world's AI chips, is short 34,000 semiconductor workers ^[7]. And the buildout is bidding up the trades it needs on site, with Meta, Amazon and Google offering skilled workers raises of 25 to 30 percent to come build data centers ^[8]. The code column keeps compressing in the meantime. Microsoft's own analysis counts 11 million Americans in desk-bound roles as highly vulnerable to AI automation, against 5.5 million in hands-on roles who are least vulnerable ^[9]. The industry that erases the first number is the same one bidding up the second. The two columns share one broken part: the bottom rung. IBM's chief talent officer put the software side of it without flinching.

Traditionally people were building judgment through actually doing the work. — Aparna P Nair

With the machine doing the work, there is no cheap seat where a beginner learns judgment by doing it. The trades carry the mirror image: roughly 400,000 skilled-trade positions sit unfilled in the United States, and the shortage predates the data center boom, the residue of decades of advice that pushed everyone toward a desk ^[10]. One ladder lost its bottom to a machine; the other never had one, because nobody was told to climb it. It is worth being precise about what the skeptics are doubting. Michael Burry is short the memory-chip trade on the bet that supply catches up and the shortage blows off ^[11]; others point to the billions in data center debt parked in special-purpose vehicles ^[12]. But both are arguments about how long today's scarcity lasts, not about where it sits. Financing can fail and the electrician shortage would still be standing ^[10]. Mike Rowe, the trades advocate, said what the whole arc adds up to.

Yeah, well, AI's coming for the coders. — Dirty Jobs

For fifteen years, this same industry told a generation the future was in code. Jensen Huang, again, this time on the recruiting trail.

You don't need to have a PhD in computer science to do so. — Jensen Huang