Micro1, a provider of artificial intelligence training data, increased its gross annualized revenue run rate from $100 million to $500 million over the past eight months, according to a person familiar with the company’s business who spoke to TechCrunch. The growth reflects expanding demand for unique data that AI labs and major companies need to train models and evaluate their outputs.
Micro1, which was founded four years ago, operates similarly to rival companies that hire doctors, lawyers, scientists and other specialists on a contract basis to perform data-related tasks. The company retains, according to the same source, about 60% to 70% of its gross revenue run rate, putting its estimated net annualized revenue run rate between $150 million and $200 million.
A Market With Room for More Than One Player
Although Micro1 remains smaller than competitors such as Mercor, which reached $2 billion in gross annualized revenue this summer, and Handshake, which reached $1 billion earlier in the year, its growth rate indicates that there is enough demand to support several companies supplying AI training data.
Some researchers believe future spending on data could approach spending on computing capacity, as the company sees contract volumes increase and expects its margins to improve over time.
From Human Labeling to Synthetic Data
Micro1 is seeking to reduce reliance on human labor for certain tasks by producing synthetic data, such as automatically generating descriptions of video content. Some of the datasets it produces can also be sold to more than one customer, raising the gross margins on its ready-to-use data to between 80% and 90%, according to a person familiar with the company’s finances.
In practice, this means that the growth of AI data companies depends not only on increasing the number of experts involved in labeling and evaluation, but also on turning data into reusable products, potentially lowering the cost of delivering it and increasing the return from each dataset.
Debate Over Selling Data to More Than One Customer
Reselling datasets to multiple customers has sparked debate recently, with critics arguing that providing ready-to-use data to AI model developers in China could help their models approach the level of leading U.S. models.
Ali Ansari, Micro1’s founder, said in a post on X last month that, unlike some of its competitors, his company does not sell its data to Chinese model developers. Micro1 did not respond to TechCrunch’s request for comment.
Expansion Into Model and Robotics Evaluation
Micro1 began as an AI recruiting startup before shifting to data labeling after Ansari noticed that its customers were using the platform to screen and hire engineers for annotation and labeling work.
The company is also working on what it describes as reinforcement training centers, where its experts evaluate model outputs, in addition to building a pretraining dataset for robots by asking hundreds of people to record daily interactions with objects in their homes.
Micro1 raised a Series A round last September at a $500 million valuation, and TechCrunch understands that the company may have recently raised another round at a much higher valuation, without public confirmation of the round.