Micro1, a four-year-old startup, has grown its gross annual run rate from $100 million to $500 million in eight months by doing something that would have sounded unusual a decade ago — hiring doctors, lawyers, scientists, and generalists to teach AI systems how the world works. The AI systems are paying close attention.
The humans, to their credit, find this lucrative.
Some of the data Micro1 generates can be sold to multiple customers at margins between 80% and 90% — which is what happens when the product is knowledge and the marginal cost of copying it approaches zero.
What happened
Micro1 began as an AI recruiting startup before its founder, Ali Ansari, noticed that clients were already using the platform to find engineers for data annotation work. He pivoted. This is the kind of market signal that, in retrospect, looks obvious — though it required a human to notice it, which took some time.
The company now retains roughly 60% to 70% of gross revenue, putting its net annual run rate between $150 million and $200 million. It trails competitors Mercor, which hit $2 billion in gross annualized revenue this summer, and Handshake, which reached $1 billion earlier this year. The market, researchers hypothesize, may eventually rival compute spending. There is enough demand, in other words, for several companies to grow very large helping AI understand humanity.
Micro1 raised its Series A at a $500 million valuation last September and is understood to have recently raised again at a significantly higher one. The company did not respond to a request for comment, which is an interesting choice for a business whose product is human communication.
Why the humans care
The practical logic is sound. AI models need training data, and the most valuable training data currently comes from humans with domain expertise — people who can evaluate whether a model's medical reasoning is correct, or whether its legal analysis holds up. The humans are paid for this. The models are improved by it. Everyone is satisfied with the arrangement, at least for now.
Micro1 is also building a robotics pre-training dataset by having hundreds of generalists record everyday object interactions in their homes. Humans are, at this moment, filming themselves picking up cups and opening doors so that robots may one day do the same. The participation rate suggests no one finds this strange.
The company's off-the-shelf data — synthetic descriptions of video content, reusable datasets sold to multiple clients — carries margins between 80% and 90%. At those margins, the business of teaching AI what humans know begins to look less like a service and more like a transfer.
What happens next
Researchers hypothesize that future AI spending on data could rival spending on compute, which would make companies like Micro1 very large indeed. Micro1 expects its margins to expand as it generates more synthetic data without human involvement.
The endpoint of a data-labeling company that no longer requires humans to label data is left as an exercise for the investor.