2026-04-28

When Contractors Stop Showing Up

The Microsoft-OpenAI divorce is being framed as a competition story. Finally, the reasoning goes, we'll get a real AI race instead of an exclusive partnership tax. More entrants, lower prices, faster innovation.

It's the wrong story entirely.

Start here: 4 terabytes of voice samples from 40,000 contractors at Mercor just got stolen. These aren't personal records—they're work product. The assets these people were paid to create, sitting unencrypted on someone's server. Now they're in the wild, ready to be scraped into training datasets, sold, duplicated, used to train competing models.

Then layer on what Google's own staff are doing: 600+ engineers signing an open letter rejecting a classified military AI contract. Not debating it. Rejecting it. These are senior people—directors, VPs—at the company that built the infrastructure to make AI work at scale. They're saying no.

Now consider the builders. Small construction firms in Australia and Canada reporting that they can't access credit, can't survive on current margins, and see no path forward. They're defaulting at 2x the rate of smaller debts. This isn't a market correction. This is structural collapse of the intermediary class that actually executes work.

The pattern is consistent: **workers and contractors are losing faith in the institutions asking them to produce**.

In AI training, that loss is material. These contractors create the labeled data, the voice samples, the reasoning traces that make frontier models work. They were already poorly compensated. They were already watching their work get scraped and redistributed without consent. Now their assets are stolen, their employers offer no protection, and the companies building the models openly court military contracts that the builders themselves find unconscionable.

What happens when the best contractors—the ones with alternatives—start walking?

Microsoft-OpenAI breaking exclusivity is supposed to unlock competition. But if the contractor supply shrinks while demand for training data explodes, you don't get competition. You get fragmentation. Large, vertically integrated players (OpenAI with its own capital, Microsoft with infrastructure) can afford to build internal labeling operations or negotiate at scale. Everyone else starves.

The market is pricing this as a bullish event—more players, more optionality, acceleration. It's missing the supply shock underneath: contractor availability and data quality are about to crater, and the breakup just made the divide between the haves and the have-nots permanent.

This isn't recession. It's disintermediation under duress. And it compounds fast once it starts.

The nightmare scenario isn't gradual margin compression. It's a sudden contractor exodus across AI training, construction, and gig labor, synchronized with a credit event in commercial real estate, all while tech firms face regulatory backlash for military work. Supply shock, demand shock, policy shock. The market has priced none of it because it all looks structural, not cyclical, and Wall Street doesn't have tools for that story yet.

When do the people who make the thing stop trusting the people selling it?

↓ DOWN48hconviction 32%
Conviction: 47% | Alignment: aligned_bearish
← OlderArchive