AI Training Data Startup Micro1 Hit $500 Million Gross Run Rate in Eight Months

Key points

• Micro1, a four-year-old data-labeling startup, grew its gross annual run rate from $100 million to $500 million in eight months as of mid-2025.
• The company retains roughly 60 to 70 percent of that figure, putting its net annual run rate between $150 million and $200 million.
• Competitor Mercor hit $2 billion in gross annualized revenue this summer; Handshake reached $1 billion earlier this year.
• Micro1 raised its Series A at a $500 million valuation in September 2024 and may have recently closed another round at a higher, unconfirmed valuation.
• Founder Ali Ansari publicly stated the company does not sell training data to Chinese AI developers.

AI companies need vast amounts of carefully checked, human-verified information to train their models. Gathering and labeling that information is a booming business, and Micro1 is one of its fastest-growing players.

First reported by TechCrunch, Micro1 grew its gross annual run rate, the amount it would earn in a year at its current pace, from $100 million to $500 million in eight months. A fivefold jump.

How does Micro1 actually make money?

The company connects AI labs and corporations with contract experts: doctors, lawyers, scientists who check and rate AI outputs, helping models learn what good answers look like. This process is called data labeling.

Micro1 keeps 60 to 70 percent of the revenue it collects, passing the rest to contractors. Its net annual run rate sits between $150 million and $200 million.

The startup is also building a higher-margin business around synthetic data, information generated automatically rather than by humans. Some of this data can be sold to multiple clients. When the same dataset ships to several buyers, the cost of creating it is shared, so margins on those "off-the-shelf" datasets can reach 80 to 90 percent, according to a person familiar with the company's finances.

One example: automated written descriptions of video content, created without a human watching every clip.

Does it matter who buys this data?

Yes, and it's already controversial. Critics have argued that selling the same datasets to Chinese AI developers helps those developers build models that rival the best American ones.

Micro1's founder Ali Ansari addressed this directly last month, saying the company does not sell its data to Chinese model makers. "Some human data companies work with foreign adversaries, and the results show today in Kimi K3," he posted on X, referring to a Chinese AI model. "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."

The controversy sits inside a broader question our 12 August story on OpenWALDO raised: when training data has no public audit trail, nobody outside the company can verify who received it.

Should you worry about where AI training data ends up?

If you work in a field whose expertise feeds these models, whether medicine, law or engineering, the answer is worth your attention. The contractors doing this work are real professionals whose judgements shape how AI systems behave for millions of people, yet the governance around who buys the resulting data remains thin.

What happens next?

Micro1 is not alone. Mercor hit $2 billion in gross annualized revenue this summer; Handshake reached $1 billion earlier this year. The market appears large enough for all of them.

Some researchers suggest future AI spending on data could eventually rival what the industry spends on compute, the hardware running AI systems. Micro1 raised its Series A at a $500 million valuation in September 2024 and may have recently closed another round at a higher valuation, though details remain unconfirmed. The company did not respond to a request for comment.