Micro1 hits $500 million run rate in eight months

Micro1 (Micro1) doubled its gross run rate fivefold in less than a year, from $100 million to $500 million, according to a source close to the company. After retaining 60% to 70% of the amount, its net run rate stands at $150 million to $200 million per year. The growth illustrates how demand for unique training data has become a bottomless pit for large model labs.

Micro1 remains far from the leaders: Mercor reached $2 billion in annual gross run rate in the summer, and Handshake crossed $1 billion earlier this year. Researchers estimate that data spending may soon equal compute spending, leaving room for several players simultaneously. Micro1’s contracts are accelerating, and the company forecasts that margins will expand over time.

A growing share of revenue comes from synthetic data, such as automatically generated video-content descriptions created without human involvement. Off-the-shelf data sold to multiple customers simultaneously yields gross margins of 80% to 90%, according to the same financial source. This model lets the company reuse the same preparation work repeatedly instead of hiring experts for each project.

Selling the same data sets to multiple customers has drawn criticism: some argue that distributing ready-made data to Chinese developers levels their models against leading U.S. offerings. Founder Ali Ansari wrote last month on X that Micro1 does not sell data to Chinese model manufacturers, unlike some competitors. “Human-data companies work with foreign adversaries. The results are visible today in Kimi K3. We believe it is disgraceful to claim American AI dominance while selling data for millions to countries we are in fierce competition with,” he wrote.

Micro1 began as an AI-driven recruiting startup, similar to Mercor. Ansari noticed that data-labeling customers used its platform to filter and recruit annotation engineers, and decided to target the labeling market itself. Beyond model-output evaluation by experts—a concept dubbed “reinforcement learning gyms”—the company is building pre-training data sets for robotics: hundreds of people record daily interactions with objects in their homes. A Series A round closed in September at a $500 million valuation, and, according to TechCrunch, a subsequent funding round may have closed recently at a substantially higher valuation. Micro1 did not respond to a request for comment.