AI training-data company Micro1 has reached a $500 million gross annual run rate, according to a person familiar with the business who spoke with TechCrunch. The figure represents a fivefold increase from $100 million in eight months and places the four-year-old startup among the fastest-growing suppliers in the model-development economy.
Gross run rate is not the same as revenue retained by the company. After payments associated with the experts it hires, TechCrunch’s source puts Micro1’s net annual run rate between roughly $150 million and $200 million. Micro1 did not comment on the figures, and the published gross-retention description does not reconcile cleanly with that net range.
Training data becomes a larger AI business
Micro1’s growth sits alongside rapid expansion at rivals. TechCrunch reports that Mercor has reached a $2 billion gross annualised run rate, while Handshake has reached $1 billion. These comparisons use gross transaction measures, so they should not be read as equivalent to conventional software revenue.
The companies benefit from growing demand for human expertise and specialised datasets used to improve AI models. One researcher cited by TechCrunch argues that future spending on data could rival spending on computing capacity. That is a projection rather than a measured market outcome, but current supplier growth shows why investors and model developers are paying attention.
Micro1 did not begin as a data-labelling company. It initially built an AI recruiting platform. Founder Ali Ansari said customers were using the service to recruit engineers who could evaluate and improve model outputs, which led the startup to pivot toward data work.
Today, experts help create evaluation environments sometimes called reinforcement-learning gyms. They assess model responses and provide signals that developers can use during training. The company is also building a robotics pretraining dataset by employing hundreds of people to record interactions with household objects.
Synthetic datasets change the economics
Micro1 is expanding into synthetic data, including automatically generated descriptions of video. Unlike a custom project created for one model developer, an off-the-shelf dataset can be sold repeatedly. The company claims this approach can produce gross margins of 80% to 90%.
Those margins are Micro1’s assertion, not audited results disclosed in the report. Their durability will depend on the cost of creating, verifying and updating datasets, as well as whether multiple customers continue to value the same material. Data quality and provenance can be as important as volume when errors may propagate into a model.
The market has also attracted scrutiny over which customers receive advanced training resources. Some data suppliers have faced controversy for selling datasets to Chinese AI developers. Ansari told TechCrunch that Micro1 does not do so. The source provides the founder’s statement but does not describe an independent audit of the policy.
Reusable data may allow suppliers to move beyond labour-intensive annotation contracts. It also creates a responsibility to document how material was collected and what permissions apply. TechCrunch’s account focuses on commercial growth and does not establish how Micro1 addresses every provenance, privacy or licensing question across its catalogue.
Fast growth still needs careful interpretation
Micro1 raised a Series A round at a $500 million valuation last September. TechCrunch reports that the company may since have raised another round at a higher valuation, although it does not confirm the size, timing or investors. The startup declined to comment on its financial performance.
The headline growth rate is therefore based on information from a person familiar with the company rather than public financial statements. Gross annual run rate extrapolates current activity and can change quickly. The lower net revenue estimate offers a more conservative view, but it too depends on the accuracy of the reported retention range.
Even with those limitations, the trajectory shows how model development is supporting a new layer of infrastructure businesses. Compute receives much of the attention, but training also requires expert judgement, evaluations and carefully assembled examples. Micro1’s challenge is to turn extraordinary transaction growth into repeatable, defensible revenue while proving that its data meets the standards increasingly expected by AI developers.
