OpenAI Is Bidding Up Electricians While DeepMind's AlphaFold Team Walks Out the Door
I’ve been dismissive of the line that AI would create more jobs than it destroys. It always sounded like something you say to a select committee — technically unfalsifiable, emotionally reassuring, and never accompanied by a name or a wage. I assumed the people saying it couldn’t produce a single worker who was better off.
They can now. It’s just nobody they were talking about.
OpenAI, Google, Meta, and BlackRock are recruiting electricians and carpenters by the thousands, at some of the highest pay the construction trades have ever seen. In the same twenty-four hours, DeepMind has reassigned most of the original AlphaFold authors over the past year, with nearly a quarter of the full-time authors gone entirely.
The team behind the most celebrated piece of computational science of the decade is dispersing. The journeyman wiring the substation that powers the next model is having the best year of his career.
Ask which worker gained the most, relative to what they had
That’s the honest test, and it isn’t close.
A frontier researcher was already among the best-compensated technical workers on earth. Moving between labs, however lucrative, is a lateral trade for someone who had options at every step. Lilian Weng announced she was leaving Thinking Machines Lab over workload and health concerns, and days later was reported returning to OpenAI. That’s a person with total mobility inside a small labor market where every employer wants her.
Now take the electrician in Kentucky, where Brookfield and NextEra just announced a $100 billion campus on a former uranium-enrichment site, more than 1.2 gigawatts, opening in 2032. Seven years of scheduled work in a county where the last big employer was a federal facility that wound down. That’s not a lateral move. That’s a wage floor lifted for a whole trade, in a place, on a timeline long enough to apprentice a kid into it.
Measured in percentage terms against their prior wage, the carpenter is the biggest winner in artificial intelligence this year.
The research is being given away, which tells you what it costs
Here’s the part that reframes the AlphaFold departures. OpenAI just launched ChatGPT for Academic Researchers, handing 100,000 scientists, mathematicians, and engineers free access to frontier models through 2027.
You don’t give away the scarce input. You give away the abundant one to build dependence on the thing that isn’t abundant, which is the compute underneath it and the power contract underneath that.
So consider what a research organization looks like on that logic. Once the paper is written and the weights exist, the team that produced it is a cost center with a high salary line and a publication incentive that pulls against product. Reassigning them isn’t a failure of institutional memory. It’s what you do when the model is the asset and the modelers were the acquisition cost.
The trades are on the other side of that trade. You cannot download a licensed electrician, you cannot fine-tune a crew that can pull cable to code, and the bottleneck on 1.2 gigawatts isn’t a research idea. It’s people who can terminate a bus duct.
Scarcity moved. Compensation followed it, exactly the way the textbook says.
What I still don’t buy
None of this vindicates the jobs argument as it’s usually made. Construction employment is cyclical and this is a build-out, not a permanent economy. Ask the fiber crews of 2001 how the boom’s second act went. A 2032 completion date is a long way from a career, and when the campuses are energized the operating headcount is a fraction of the building headcount.
And the dispersal of a team like AlphaFold’s is a real loss that no wage number offsets. That group solved a fifty-year problem. Whatever the institutional logic for reassigning them, a research culture that treats its highest achievement as a completed line item will not produce the next one.
The carpenter’s raise is real. So is the empty desk in London.
Both are the same decision.