Micro1’s Reported $500M Run Rate Tests the Economics of AI Training Data

Micro1’s gross annual run rate reportedly rose from $100 million to $500 million in eight months. The increase puts a price on the demand for specialized material used to train and evaluate AI models, but it also leaves a basic question about how much of that reported gross volume the company retains.

Micro1 began as an AI recruiting startup before founder Ali Ansari moved it into data labeling. The shift followed clients’ use of its recruiting platform to vet and hire engineers for annotation work. The company now draws on contract domain specialists, including doctors, lawyers and scientists, for training-data work.

The $500 million number is a gross annual run rate, supplied by a person familiar with Micro1; the company did not respond to TechCrunch’s request for comment. TechCrunch also wrote that Micro1 retains roughly 60% to 70% of gross run rate, while putting its net annual run rate at about $150 million to $200 million.

Applied to a $500 million gross run rate, retaining 60% to 70% would equal $300 million to $350 million, not $150 million to $200 million. TechCrunch did not reconcile those figures. The discrepancy means the reported retention percentage and net-run-rate estimate cannot both be used as one calculation of Micro1’s economics.

Micro1 is increasingly generating synthetic data without human involvement, including automated descriptions of video content. Some datasets can be sold to multiple customers instead of being made for a single engagement. A person familiar with its finances said this off-the-shelf data can produce gross margins of 80% to 90%.

Two sources of training material

• Experts evaluate model outputs in a process known as reinforcement-learning gyms.
• Hundreds of generalists record everyday interactions with objects in their homes for a robotics pre-training dataset.

Reselling a dataset can improve its economics, but the practice has drawn criticism when off-the-shelf AI data goes to Chinese developers. Critics argue such sales can help Chinese models approach leading U.S. models. Ansari said last month that Micro1 does not sell data to Chinese model makers; that is the founder’s stated policy, not an independently verified customer list.

Micro1 remains smaller than Mercor, which TechCrunch said reached $2 billion in gross annualized revenue this summer, and Handshake, which it said reached $1 billion earlier this year. Micro1 raised its Series A at a $500 million valuation last September. TechCrunch says it may have raised another round at a significantly higher valuation, but Micro1 has not confirmed that financing or its terms.