Snorkel AI, a San Francisco startup that supplies training data and reinforcement-learning environments to artificial-intelligence developers, said Tuesday it had raised $350 million in a Series E round that values the company at $3.5 billion, nearly triple the $1.3 billion valuation it carried when it last raised money 17 months ago.
The round was co-led by Insight Partners and Section 32, with returning backers Addition, Lightspeed, Greylock, GV and Wells Fargo also participating. Snorkel said its annualized revenue run rate has crossed $350 million, up from roughly $20 million a year ago, a surge the company attributes to a data-as-a-service business it launched last year.
Founded in 2019 by researchers from Stanford's AI lab, including chief executive Alex Ratner, Snorkel initially sold software that automated the labeling of training data for machine-learning teams. It has since shifted toward delivering finished datasets directly to AI labs, blending synthetic data generation with work from human subject-matter experts, an approach the company calls its data factory.
The teams pushing the frontier want a research data partner who pioneers the science of data development. That's what Snorkel was built to be: the frontier lab for agentic data, combining human excellence with over a decade of research and technology.
Alex Ratner, co-founder and CEO, Snorkel AI
A crowded but lucrative data race
The raise is the latest sign that venture investors remain willing to write large checks for the unglamorous infrastructure layer beneath frontier AI models, even as public markets debate whether spending on AI capacity is outrunning near-term returns. It follows a string of similar deals this week, including a $75 million Series B for web-data infrastructure startup Firecrawl and a combined $340 million equity-and-credit package for clinical AI company Heidi Health.
Much of that appetite reflects a shift in what AI labs need most. As models move from answering questions to taking multi-step actions on a user's behalf, generic text scraped from the web is no longer enough; labs increasingly need curated examples and simulated environments that teach a model how to reason through a task and recover from mistakes, work that leans heavily on subject-matter experts in fields such as law, medicine and software engineering rather than general-purpose crowdworkers.
Snorkel said the new capital will fund expansion of its data factory and additional hiring of researchers as it competes for contracts with the handful of labs building the most advanced general-purpose models, a market that has grown more crowded, and more lucrative, as demand for complex, expert-vetted training data outpaces the supply of engineers able to produce it.