Micro1, a four-year-old startup supplying human and synthetic data for AI training, has expanded its gross annual run rate from $100 million to $500 million over the past eight months, according to a person familiar with the company.
Like other companies in the sector that recruit domain specialists such as doctors, lawyers, and scientists on a contract basis, Micro1 retains roughly 60% to 70% of its gross annual run rate. That puts the company’s net annual run rate between $150 million and $200 million.
Despite its rapid expansion, Micro1 remains behind some of its larger competitors. Mercor reached $2 billion in gross annualized revenue this summer, while Handshake crossed $1 billion earlier this year.
Still, Micro1’s growth indicates that demand for companies supplying data and human expertise to train AI models may be large enough to support several major players in the market.
The opportunity could expand further as AI companies increase their spending on training data. Some researchers have hypothesized that future AI spending on data could eventually rival expenditure on compute.
That trend could benefit Micro1 as the startup sees contract sizes increasing at an accelerated pace and anticipates its margins improving over time.
One factor behind the expected margin expansion is Micro1’s increasing use of synthetic data that does not require human involvement. The company, for instance, creates automated descriptions of video content.
Some of the data generated by Micro1 can also be sold to multiple customers. As a result, gross margins on this “off-the-shelf” data can reach as high as 80% to 90%, according to a person familiar with the startup’s finances who spoke to TechCrunch.
The practice of selling identical datasets to multiple clients has nevertheless attracted controversy. Critics have argued that providing off-the-shelf data to Chinese AI developers could help those companies build models that are as capable as leading U.S. systems.
Micro1 founder Ali Ansari said last month on X that the company does not sell its data to Chinese model makers, distinguishing Micro1 from some competitors.
“Some human data companies work with foreign adversaries, and the results show today in Kimi K3. We believe it’s shameful to claim American AI dominance desires while selling millions worth of data to countries that we are in adversarial competition with.”
Micro1’s origins are also closely tied to AI recruitment. Like Mercor, the company initially operated as an AI recruiting startup.
Ansari later noticed that customers using Micro1’s data-labeling services were also using the company’s AI platform to assess and recruit engineers for annotation work. That prompted him to expand Micro1 into the data-labeling business.
The company’s work extends beyond conventional data annotation. Ansari previously said that Micro1 uses its experts to evaluate AI model outputs, a process known as reinforcement learning gyms.
The startup is also developing a robotics pre-training dataset by having hundreds of generalists record everyday interactions with objects in their homes.
Micro1 raised its Series A at a $500 million valuation last September. The startup may have recently raised another round at a significantly higher valuation, according to information understood about the company.
Micro1’s latest growth comes as the AI industry continues to increase its reliance on specialised human expertise, synthetic datasets, and other forms of training data. The company’s expanding use of automated data generation could also provide a path toward stronger margins as it scales.
With its gross annual run rate now at $500 million, Micro1 is rapidly closing the gap with established AI data providers while expanding into synthetic and robotics-related datasets. The startup’s potential new funding round at a significantly higher valuation, combined with growing contract sizes and higher-margin data products, points to continued expansion as demand for AI training data accelerates.
