Pomegra
Snorkel AI closed a $350 million Series E at a $3.5 billion valuation, tripling its worth in 17 months as frontier AI labs compete for complex training data they can no longer build themselves.
Key Takeaways
- Snorkel AI raised $350M in a Series E co-led by Insight Partners and S32, announced September 22, 2026.
- Valuation hit $3.5B, nearly triple the $1.3B it carried after its $100M Series D in April 2025.
- Annualized revenue run rate reached $375M, up from roughly $20M a year earlier - an 18x jump.
Lead
Snorkel AI, the San Francisco-based training data company, closed a $350 million Series E on September 22, 2026, valuing the seven-year-old startup at $3.5 billion. Insight Partners and S32 co-led the round. Existing backers Addition, Lightspeed, Greylock, GV, and Wells Fargo participated alongside new entrants March Capital, Blumberg Capital, Allegis Capital, Frontline, Standard, and Third Point Ventures. The raise arrives barely 17 months after Snorkel last went to market, when a $100 million Series D put it at $1.3 billion. The valuation math alone describes an unusual trajectory.
What Does Snorkel AI Actually Sell?
The short answer today is different from the answer in 2024. Snorkel began as a developer of software that automated data labeling - tools that enterprises and researchers ran internally. In September 2025, the company pivoted its primary commercial offering to data-as-a-service: customers receive completed, expert-constructed datasets rather than a platform to build their own.
The distinction matters commercially. Frontier AI labs - the handful of organizations training the largest models - have hit a ceiling on the data they can generate with general-purpose labelers. The tasks required to push the next generation of reasoning and agentic systems demand subject matter experts who may spend hours or days constructing a single example. Snorkel's model layers its own software and synthetic data generation on top of those experts, then delivers finished training sets, benchmarks, evaluations, and reinforcement learning environments as contracted outputs.
That shift from software vendor to managed service fundamentally changes the revenue model - and the addressable customer base.
Why Did Revenue Grow 18x in One Year?
The data-as-a-service launch in September 2025 is the direct answer. Snorkel's annualized run rate stood near $20 million before that pivot. By September 2026, it reported a run rate of $375 million. No single product release produces that curve without a structural change in what is being sold and who is paying for it.
The structural change here is the rising cost of frontier training data. Large language model developers and their successors increasingly outsource the expert data creation they cannot staff internally. Snorkel is positioned as an industrial supplier to that process - not a tool sold to the labs' data teams, but a factory that delivers outputs on contract. The reinforcement learning environments it now sells extend that positioning further into the post-training and evaluation workflows that have grown in importance as base model training itself scales.
The 18x ARR number is the kind of metric that either describes genuine product-market fit or, less charitably, reflects a handful of very large contracts that may not recur. At $375 million in run rate and $3.5 billion in valuation, the implied multiple is roughly 9x forward revenue - elevated, but not aberrant for a category that major investors currently view as critical infrastructure.
Valuation Trajectory and What the Round Implies
The Series D valued Snorkel at $1.3 billion in April 2025. The Series E at $3.5 billion represents a 2.7x step-up in under a year and a half. For investors in the earlier round, the markup is attractive. For the Series E investors, the implied bet is that $375 million in ARR is a floor, not a ceiling.
The round also signals continued LP appetite for picks-and-shovels AI infrastructure plays. Rather than betting on which frontier model wins, training data suppliers benefit from competition among labs - more models in training means more demand for diverse, expert-grade data. That logic has driven significant capital into the sector across 2025 and 2026.
Outlook
Snorkel enters the second half of 2026 with a clear thesis: that frontier AI development has created a sustained industrial demand for expert training data that no single lab will satisfy internally. Its pivot to data-as-a-service has been commercially validated at speed. The questions now are whether the $375 million run rate is durable as lab spending priorities shift, whether synthetic data generation eventually reduces the need for human expert inputs that justify Snorkel's pricing, and how many well-capitalized competitors will attempt the same model. The $350 million gives the company runway to answer those questions without pressure.
