The Enigma of the Data Labeling Startup Explosion

If you’re like me, you’ve been watching the explosion of data labeling and reinforcement learning environment startups with a mix of confusion and awe. 

These companies, which hire experts in fields like medicine, law and software engineering and build fake copies of apps like Salesforce and Excel to produce training data for AI models, are https://www.theinformation.com/articles/handshake-mercor-revenue-surges-demand-human-contractors-train-ai">rapidly growing both in terms of revenue and valuations.

But their moment in the sun may soon be about to pass.

These startups seem to lack any sort of long-term defensible moat.

Their contracts with the big AI labs varies—some labs, such as Anthropic, have tried lots of different startups, researchers say.

Individual teams or individual researchers often strike their own contracts with data labeling startups.

At other firms, such as OpenAI, a centralized data team manages relationships with data labeling firms.

These smaller contracts can range anywhere from hundreds of thousands to millions of dollars.

That means data startups can fairly quickly get from zero to $10 million in annualized revenue, researchers tell me.