Snorkel AI raises $350 million for AI data

The company stopped licensing its platform to deliver training data and environments instead, and the round values it at $3.5 billion.

What happened

On September 22, Snorkel AI announced a $350 million round that values it at $3.5 billion, according to the press release distributed by the company itself.
The round is led by Insight Partners and S32. March Capital, Blumberg Capital, Allegis Capital, Frontline, Standard and Third Point Ventures come in, and Greylock, Lightspeed, GV, Addition, Factory, Prosperity7, Walden Catalyst and Wells Fargo return.
The stated use of the funds is to expand the capacity of its data factory for agents, accelerate investments in vertical and enterprise AI, and extend research to new domains and modalities.
Its chief executive, Alex Ratner, describes the company as “the frontier lab for agentic data.” The press release doesn’t report recurring revenue or the names of buyers.

The shift is the story, not the amount. Snorkel was born licensing a labeling platform and today charges for delivering datasets and training environments. The software stopped being the product.
For teams in the region evaluating whether to build their own AI capabilities, this reorders priorities: the differentiator is no longer in the tool you buy, but in the material you feed it. And that material, in River Plate, Chilean or Andean Spanish, doesn’t exist in packaged form.
A consequence the press release doesn’t mention: if training data is sold by domain and by language, every organization with well-organized archives of its own is sitting on an input that now has a market price. Most don’t know it, and they hand it over for free with every integration.

$3.5 billion. The valuation the round assigns to a company that sells data, not licenses.

The move runs in parallel to what we’ve already noted about how the big tech companies account for AI in their results and about Mistral’s free plan, which is paid for with training data. Three different ways of recognizing that the bottleneck is no longer the model.

No timelines announced for deploying the additional capacity or for publishing revenue metrics.
The company didn’t say whether it will open direct operations in Latin America.

For a decade the promise was that data was the new oil. Only now is someone charging like a refinery.