Surging demand for unique training data from leading AI labs and companies is fueling strong growth for startups that label and supply datasets. One such fast‑growing firm is Micro1, a four‑year‑old company that increased its gross annual run rate (ARR) from $100 million to $500 million over the past eight months, according to a person familiar with the company.
Net revenue and margins
That source said Micro1, like peers that contract domain experts such as doctors, lawyers and scientists, retains roughly 60–70% of its gross revenue. Based on that share, the company’s net annual run rate is estimated between $150 million and $200 million.
Market context and competitors
Micro1 still trails larger rivals—Mercor reached $2 billion in gross annualized revenue this summer, and Handshake hit $1 billion earlier this year—but Micro1’s revenue surge suggests there is room for multiple substantial suppliers of training data. Some researchers hypothesize that future AI spending on data could rival spending on compute, a trend that would further support growth in the data supply segment.
Product strategy: synthetic and off‑the‑shelf datasets
Micro1 is increasingly producing synthetic data without human involvement, for example by generating automated video descriptions. A person familiar with Micro1’s finances told TechCrunch that datasets the company can sell to multiple customers—so‑called “off‑the‑shelf” data—can carry gross margins as high as 80–90%.
Selling the same datasets repeatedly has drawn criticism, with opponents arguing that distributing off‑the‑shelf data to Chinese AI developers can help their models approach the capabilities of top U.S. models.
Company stance and statements
Micro1 founder Ali Ansari wrote on X last month that, unlike some competitors, the company does not sell data to Chinese model makers. Ansari said: “Some human data companies work with foreign adversaries. [A]nd 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.”
Origins and other initiatives
Like Mercor, Micro1 began as an AI recruiting startup. After noticing that clients were using the company’s AI platform to vet and recruit annotators, Ansari pivoted the business into data labeling as well. He has previously told TechCrunch that beyond having experts evaluate model outputs—an approach related to reinforcement learning gyms—the company is building a robotics pre‑training dataset by having hundreds of generalists record everyday object interactions in their homes.
Funding and response
Micro1 raised its Series A at a $500 million valuation last September; TechCrunch understands the startup may have recently closed another round at a significantly higher valuation. Micro1 did not respond to a request for comment.
Why this matters
Rising demand for training datasets and the emergence of high‑margin, reusable synthetic data are reshaping the business models of AI‑training services. Micro1’s rapid growth indicates the market can support multiple major providers, while the geopolitical and ethical questions about dataset distribution—particularly sales to foreign developers—are likely to prompt further scrutiny and debate.
