The data factory for frontier AI — programmatic labeling born at Stanford, now delivering expert datasets and RL environments at $375M ARR
HQ: Palo Alto / Redwood City, California, United States | Founded: 2019 | Employees: Not publicly disclosed (hundreds) | Stage: Series E ($3.5B valuation) | Website: https://snorkel.ai
Snorkel AI grew out of one of the most influential data-centric AI research programs of the last decade. Christopher Ré's Stanford lab created Snorkel, a system for programmatic labeling: instead of hand-labeling millions of examples, domain experts write labeling functions and the system aggregates them into massive weakly-supervised training sets. Alex Ratner, then a PhD student in the lab, commercialized the research — Snorkel AI launched in 2019 with Ré as co-founder.
The company's first act sold the software: enterprises used Snorkel's platform to label training data programmatically for their own models. A $85M Series C (2024, at a $1B valuation) and a $100M Series D (April 2025, $1.3B, led by Addition) funded expansion as enterprises adopted LLMs and needed evaluation and fine-tuning data.
The second act is the one that matters. As frontier labs shifted to reinforcement learning and expert-driven post-training, the bottleneck moved from cheap crowd labels to scarce domain expertise — radiologists, lawyers, financial analysts, security researchers. In 2025 Snorkel pivoted to data-as-a-service: delivering complete datasets and RL environments, produced by a hybrid of its software, synthetic generation, and networks of subject-matter experts. The commercial result was extraordinary — a $375M annualized revenue run rate by September 2026, an eighteenfold increase over twelve months, driven by AI labs' insatiable demand.
The $350M Series E (September 2026), led by Insight Partners and S32 with Third Point, March Capital, Addition, Lightspeed, Greylock, GV, and Wells Fargo participating, nearly tripled the valuation to $3.5B. Unlike pure expert-marketplace rivals (Mercor, Handshake, Micro1 — whose gross revenue largely flows to specialists), Snorkel's expert payments sit in cost of goods sold because it sells finished datasets and environments rather than human labor. That distinction — product vs. marketplace — is the company's core strategic argument as the 'data factory' era of AI consolidates.
Snorkel spotted that frontier AI's bottleneck moved from compute to high-quality expert data — and pivoted from selling labeling software to delivering finished datasets and RL environments at $375M ARR, an 18x jump in a year. Its $3.5B valuation reflects the new reality: AI labs pay enormous premiums for the data that makes models measurably smarter.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki