Invest1 publisher2 min readPublished
Snorkel sells about a tenth of itself for $350 million at a $3.5 billion valuation
Insight Partners and S32 co-led the round, and the product they are funding is expert-built agentic environments and rubrics that Snorkel says take qualified humans hours or days to construct.
The Investor · Invest desk
What happened
- Snorkel AI said on September 22, 2026 that it had raised $350 million at a $3.5 billion valuation, with existing investor Addition joining the two co-leads in significant size.
- Thirteen further named funds took part, including new investors March Capital, Blumberg Capital, Allegis Capital, Frontline, Standard and Third Point Ventures and existing backers Greylock, Lightspeed, GV and Wells Fargo.
- The money is earmarked for capacity in the agentic data factory, vertical and enterprise AI, new domains and modalities, and open research including the company's Open Benchmarks Grants programme.
- Snorkel launched the expert Data-as-a-Service offering that sells this work in September 2025 and says it has grown rapidly since then.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint The announcement does not include revenue, so the $3.5 billion cannot be tested against a multiple, and the next buyer of this equity is pricing the investor list and the demand story.
- cost Capacity here is expert hours: if one environment consumes hours or days of a qualified specialist, growing the factory means paying for more of those people, and the cost rises with volume sold.
- exposure Snorkel describes its customers only as leading AI labs and enterprises, so anyone pricing this round is taking revenue concentration on trust.
- precedent Funding open benchmark grants while selling evaluations puts a vendor in the business of shaping the measurement standard its own product is bought against.
The check buys about a tenth of the company. $350 million against $3.5 billion is 10 percent if that valuation is post-money and 9.1 percent of $3.85 billion if it is pre-money. The announcement does not say which [1][15][1].
For Insight Partners, which reported over $90 billion in regulatory assets under management as of December 31, 2025 [12], the entire round comes to 0.39 percent of that base [3]. Insight co-led with S32, and thirteen further named funds took part [1][2][6].
Snorkel's research record is longer than its commercial one. More than 250 peer-reviewed papers, cited more than 25,000 times, about 100 citations a paper [7][2]. Against that, twelve months of selling the expert Data-as-a-Service line it launched in September 2025 [5][5]. The release dates the company's origin twice, at the Stanford AI Lab "nearly a decade ago" and, lower down, at 2019, which from September 2026 is seven years [6][4].
The demand argument is Snorkel's own. The release describes a Data 1.0 era of simple labelling, "a volume problem solved with headcount". The Data 2.0 present is expert agentic tasks, environments and rubrics that "take even the most qualified humans hours or days to construct" [4]. Designing them well, it says, is research work, where quality and complexity determine value [4].
"The teams pushing the frontier want a research data partner who pioneers the science of data development," said Alex Ratner, Snorkel's co-founder and chief executive [8]. Lonne Jaffe is a managing director at Insight. He said the firm was backing the team "as they continue to bring their frontier AI data factory capabilities to longer-running, complex work in high value industries like healthcare, law, and software engineering" [9]. Andy Harrison of S32 said the expert-agentic environments "create a unique flywheel between human expertise and AI, delivering data at the speed and quality that enables a new era of efficiency and accuracy for model development" [10].
I would underwrite the healthcare, law and software engineering work before the lab work, because enterprise buyers would still be paying if frontier training consolidates into fewer hands. Two other readings hold up. The labs keep buying expert hours, and Snorkel compounds on a cost base that grows alongside them. Or the labs assemble their own expert pools and generate environments with their own models. Then the price per built environment falls, and $3.5 billion was a price set at the top of the demand. What would settle it is a revenue disclosure that separates enterprise customers from lab contracts.
What to watch
- A disclosed revenue figure or a named customer. Either would let the $3.5 billion be measured against something other than the round itself.
- Whether the Open Benchmarks Grants money produces benchmarks that Snorkel's own paid evaluation work is then scored on.
- Evidence that a frontier lab has built its own expert-environment pipeline. That would shrink the buyer pool Snorkel is adding capacity for.