Snorkel AI raises $350M at $3.5B: data work is the moat
Snorkel's revenue grew 18x to a $375M run rate by selling finished datasets, not labeling software. The lesson for small studios is about what you sell.
A company that gave up on selling software and started selling the work tripled its valuation in 17 months. On September 22, Snorkel AI announced a $350 million Series E at a $3.5 billion valuation, led by Insight Partners and S32. The training data story is real, but the part worth studying if you run a services business is the pivot underneath it.
What actually happened
Per TechCrunch, the round values the seven-year-old company at nearly triple the $1.3 billion it carried after a $100 million Series D 17 months ago. Existing backers Addition, Lightspeed, Greylock, GV and Wells Fargo joined.
The growth number behind it: revenue up 18x over twelve months, to a $375 million annualized run rate.
What changed is the product. Snorkel started as data-labeling automation software — you bought the tool, your team did the work. It now sells data-as-a-service: finished datasets and reinforcement learning environments, produced with a hybrid of synthetic data generation and subject matter experts. Customers stopped buying a labeling platform and started buying labeled data.
Same underlying capability. Completely different thing on the invoice.
Why the data-as-a-service shift matters for your business
This is the clearest recent example of a pattern that applies directly to anyone selling technical services to small businesses.
Selling the tool caps you at the customer's willingness to operate it. Snorkel's software required a customer with an ML team, a data strategy, and the patience to run labeling functions. That's a small market, and every deal carries an implementation risk you don't control. Selling the finished dataset removes the customer's competence from the equation. The market gets bigger and the deal gets simpler.
We see the same failure mode constantly in small-business software: a studio builds a client a beautiful automation dashboard, hands over the keys, and six months later nothing runs because nobody on the client's staff owns it. The dashboard wasn't wrong. The delivery model was.
The practical version for a five-person shop: stop quoting "we'll build you an agent that processes invoices." Quote "your invoices get processed." Charge for the outcome, own the runtime, keep the thing working. Your margin improves because you're not rebuilding the same intake pipeline for every client, and your churn drops because the value doesn't depend on the client's staffing.
The caveat, honestly: outcome delivery means you hold the operational burden. If the model drifts, the API changes, or the vendor deprecates an endpoint at 2am, that's yours now. Price for it. The reason Snorkel can charge for finished data is that it has built the factory to produce it repeatedly — not because it's doing bespoke work faster.
And the market signal: an 18x revenue jump in expert-produced training data says frontier labs have exhausted the cheap web corpus and are paying real money for domain expertise. If you operate in a specialized vertical, the process knowledge in your business has a price now. That cuts both ways — check what your vendor contracts say about using your data for training.
Key takeaways
- Snorkel AI raised $350M Series E at a $3.5B valuation on September 22, led by Insight Partners and S32
- Revenue grew 18x in twelve months to a $375M annualized run rate
- Prior round: $100M Series D at $1.3B, 17 months earlier
- The pivot was from data-labeling software to data-as-a-service — finished datasets and RL environments, built with synthetic generation plus subject matter experts
- Selling the tool caps you at the customer's ability to operate it; selling the outcome removes that ceiling
- Outcome pricing means you own the runtime and the 2am API deprecation — price accordingly
- Expert training data commanding this much revenue means your vertical process knowledge has a market value; check what your vendor terms say about training on your data
We don't hand you a dashboard and wish you luck. Rush Commerce builds automation we run — outcome-priced, monitored, and fixed when a vendor breaks something at 2am. See what that looks like or run the numbers on your own process.
Sources: Snorkel AI, TechCrunch.
- #ai-training-data
- #funding
- #ai-services
- #productization
- #llm
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
AI CEOs brief the UN Security Council: what changes for you
France convened the Security Council's first session on AI safety risks on Sept 23, with Altman, Amodei, Bengio and Delangue briefing. Here's the operator read.
Read itOpenAI opens training-phase safety evals to outsiders
OpenAI says third parties can now assess models during training, not just before launch. No partners named, no access terms set — here's what to ask your AI vendors.
Read it