AI data startup Micro1 quintuples revenue in eight months

Micro1, a four-year-old AI data-labeling startup, grew its gross annual run rate from $100 million to $500 million over the past eight months, TechCrunch reported, citing a person familiar with the company. Like its peers, Micro1 hires domain experts such as doctors, lawyers, and scientists on a contract basis and keeps roughly 60% to 70% of what it bills, putting its net annual run rate closer to $150 million to $200 million.

Micro1 still trails the category's bigger names: Mercor hit $2 billion in gross annualized revenue this summer, and Handshake reached $1 billion earlier in the year. Micro1's own growth curve, though, points to a market with room for more than one winner, and some researchers now argue future AI spending on training data could eventually rival spending on compute itself.

Part of that growth comes from moving beyond one-off, human-labeled projects. Micro1 is increasingly generating synthetic data without a human in the loop, including automated descriptions of video content, and some of what it produces gets sold as reusable, off-the-shelf datasets to multiple customers at once. A person familiar with the startup's finances told TechCrunch that segment carries gross margins as high as 80% to 90%, well above what one-off contract labeling typically returns.

• Gross annual run rate: $100 million to $500 million in 8 months
• Estimated net annual run rate: $150 million to $200 million
• Mercor's gross annualized revenue: $2 billion
• Handshake's gross annualized revenue: $1 billion
• Off-the-shelf data gross margins: 80% to 90%

“Some human data companies work with foreign adversaries. And the results show today in Kimi K3. We believe it's shameful to claim American AI dominance desires while selling millions worth of data to countries that we are in adversarial competition with.”

— Ali Ansari, founder, Micro1, on X

That reusable-dataset model has pulled Micro1 into a live dispute. Selling the same off-the-shelf data to multiple clients, including Chinese AI developers, has drawn criticism that the practice helps those labs close the gap with top US models. Ansari has said Micro1 doesn't sell to Chinese model makers, distancing the company from rivals he accuses of doing exactly that.

Micro1 started as an AI recruiting tool, the same origin story as Mercor, and pivoted into data labeling after Ansari noticed clients using the recruiting platform to vet and hire annotators rather than engineers. Beyond having contracted experts grade model outputs, a setup known as reinforcement learning gyms, the company is building a robotics pre-training dataset by paying hundreds of generalist workers to record everyday object interactions inside their own homes, the exact kind of training material a company like Generalist AI needs to teach a robot a task from one demo. Micro1 raised its Series A at a $500 million valuation last September, and TechCrunch reports it may have recently closed another round at a markedly higher valuation, though the company didn't respond to a request for comment.

The tension inside Micro1's growth is the same one running through the whole data-labeling category: the traits that make a dataset profitable, reusability and scale, are the same traits that make it hard to control where it ends up once it's sold. A one-off, human-labeled contract for a single client stays contained by definition. An off-the-shelf dataset licensed to whoever pays for it doesn't, and a founder's public promise not to sell to a given buyer is a business policy rather than a technical barrier, one that a competitor or a future version of the same company could reverse without anyone outside the deal knowing.

This piece is informational, not a recommendation to buy, sell, or hold any asset.