What happened
Snorkel AI, which builds training datasets and simulated environments for AI labs and big companies, just closed a $350 million Series E that values it at $3.5 billion. Insight Partners and S32 led the round, per TechCrunch.
That valuation is nearly triple the $1.3 billion it commanded 17 months ago, when it raised $100 million in a Series D. Existing backers Addition, Lightspeed, Greylock, GV and Wells Fargo came back for more.
Snorkel also says that run rate is an eighteenfold jump over the past 12 months. That is a vendor claim, so treat it as one, but it explains the investor enthusiasm.
The business changed shape along the way. Snorkel started out selling software that automated data labeling. Last year it pivoted to handing customers finished datasets, a service it calls data-as-a-service.
It is not a pure marketplace of human experts, either. Snorkel mixes its own software and models, which generate data synthetically, with subject matter experts.
Insight Partners describes the core product as an "agentic data factory". In its telling, AI tooling plus a managed expert network builds the complex training data and reinforcement learning environments that frontier labs now want.
Why it matters
Everyone talks about chips and models. The quieter bottleneck is what goes into them.
Insight Partners argues that as AI takes on longer, higher-stakes jobs, labs and enterprises need less generic data and more specialized data, environments and evaluations. Much of that has to be built from scratch, not scraped.
That is the bet behind this round. If agents are going to work through multi-step tasks in specialized fields, someone has to build realistic practice worlds for them first.
Snorkel's customers include frontier AI labs, emerging research labs, AI-native apps and large enterprises.
Snorkel is not alone. TechCrunch notes that other companies positioning themselves as AI data labs have seen similar explosive growth.
The pattern is simple: the models took the week, and the people supplying their study material are sending invoices.
What to do with it
- Watch the data layer.Funding rounds like this show where labs are spending, so track training data and RL environment vendors alongside model news.
- Treat revenue claims as claims.The $375 million run rate and eighteenfold growth come from Snorkel itself, so wait for independent confirmation before quoting them as fact.
- Audit your own data.If you plan to build agents for a specialized domain, ask whether you have the expert-grade data and evaluations to train and test them.
- Consider the expert angle.Snorkel pairs software with subject matter experts, which suggests deep domain knowledge is a product, not just a job description.
Ciao, sapiens. The robots are studying hard, and the tutors just got a raise.


