The demand for high-quality training data from top AI labs and enterprises continues to rise, driving rapid growth in data labeling startups. Micro1, which was founded four years ago, has seen its annualized revenue increase from $100 million to $500 million over the past eight months, according to insiders. The company typically retains 60% to 70% of contract revenue, resulting in an estimated annual net income of $150 million to $200 million.

Currently, Micro1's size is still smaller than Mercor and Handshake, which had annualized total revenues of $2 billion and $1 billion this year, respectively. However, Micro1's growth shows that the AI training data market is forming a huge demand capable of supporting multiple companies. Some researchers even expect that future spending on data for AI may reach the same scale as computing power.

Micro1 is expanding synthetic data production by automatically generating video descriptions, and some data can be resold to multiple customers, achieving a gross margin of 80% to 90% for "off-the-shelf" data. However, reselling training data has sparked controversy, with concerns that such data might flow to Chinese AI developers, potentially enhancing their model capabilities. Ali Ansari, founder of Micro1, stated that the company will not sell data to Chinese model developers.

Micro1 originally started as an AI recruitment company but shifted to training data business after discovering that customers were using its platform to screen data labelers. Currently, the company not only has experts evaluate model outputs and conduct reinforcement learning training, but it is also building a robot pre-training dataset, where hundreds of ordinary users record their interactions with everyday objects.

In September 2025, Micro1 completed its Series A funding round, reaching a valuation of $500 million. According to reports, the company may complete another round of funding soon, with its valuation possibly rising significantly.