Snorkel AI, a startup that develops training datasets and simulated environments for AI labs and companies, raised $350 million in a Series E round led by Insight Partners and S32. The seven-year-old company’s valuation rose to $3.5B, nearly tripling from the $1.3B level it reached in its Series D round 17 months ago. Addition, Lightspeed, Greylock, GV and Wells Fargo also participated in the round.
While Snorkel initially offered data-labeling automation software, it shifted last year toward completed datasets, which it calls “data-as-a-service.” The company combines subject-matter experts with software and models in a hybrid approach that also includes synthetic data generation. Snorkel says its annualized revenue increased eighteenfold over the last 12 months to $375 million, attributing the growth to AI labs’ demand for advanced training data.
The gross annualized revenue figures reported for similar companies are as follows:
- Mercor: $2 billion
- Handshake: $1 billion
- Micro1: $500 million
Because these companies pay approximately 60% to 70% of their revenue to the experts doing the work, their net annual revenue is significantly lower than these gross figures. Snorkel, meanwhile, says it sells reinforcement learning (RL) environments and completed datasets instead of human labor, and that payments to experts are recorded as the cost of goods sold rather than as revenue.
The company began commercial operations in 2019, following four years of research by Alex Ratner and the Stanford AI lab team.
Why it matters
Snorkel’s shift in its business model from labeling automation to completed datasets and simulated environments shows that teams developing AI are seeking training infrastructure that is directly usable and incorporates expertise, rather than just raw data. The hybrid structure combines the contributions of subject-matter experts with software, models, and synthetic data generation, pointing to how the data preparation process is being commercialized. However, companies in the sector do not generate revenue in the same way: some report payments to experts as gross revenue, while Snorkel records them as the cost of goods sold; therefore, the reported figures cannot be compared directly. The open question is how this model will maintain revenue growth and data quality while meeting demand for advanced data, and how the cost structure of data services will become standardized across the sector.