Mercor's $20B Valuation: The Real Asset May Be the Verifier, Not the Data.
Mercor raised $750M at a $20B valuation. Its founder says revenue run rate reached $3B in 27 months.
Impressive. But two distinctions matter.
First, gross revenue isn't software revenue. Historical reporting suggests 60–70% of Mercor's top line went to expert contractors. That makes its economics very different from a high-margin SaaS business.
Second, Mercor's acquisition of Deeptune points to a more interesting business than labeling.
An AI agent needs a realistic environment, a task to complete, and a verifier that can determine whether it actually succeeded.
The same environment can potentially generate training feedback across thousands of attempts. That's a different economic model from paying an expert for each individual label.
But it creates a new competitive question: who owns the verifier?
If enterprises can build reliable task environments and grading systems internally, third-party margins may compress. If maintaining them requires scarce expertise and constantly updated workflows, providers like Mercor could develop a durable advantage.
My view: the next AI training moat won't be the largest contractor network. It will be the hardest-to-replicate system for measuring whether agents can do economically valuable work.
Watch enterprise revenue mix, contractor-adjusted margins, and repeat spending on evaluation environments.
#stocks
Mercor raised $750M at a $20B valuation. Its founder says revenue run rate reached $3B in 27 months.
Impressive. But two distinctions matter.
First, gross revenue isn't software revenue. Historical reporting suggests 60–70% of Mercor's top line went to expert contractors. That makes its economics very different from a high-margin SaaS business.
Second, Mercor's acquisition of Deeptune points to a more interesting business than labeling.
An AI agent needs a realistic environment, a task to complete, and a verifier that can determine whether it actually succeeded.
The same environment can potentially generate training feedback across thousands of attempts. That's a different economic model from paying an expert for each individual label.
But it creates a new competitive question: who owns the verifier?
If enterprises can build reliable task environments and grading systems internally, third-party margins may compress. If maintaining them requires scarce expertise and constantly updated workflows, providers like Mercor could develop a durable advantage.
My view: the next AI training moat won't be the largest contractor network. It will be the hardest-to-replicate system for measuring whether agents can do economically valuable work.
Watch enterprise revenue mix, contractor-adjusted margins, and repeat spending on evaluation environments.
#stocks