- Snorkel AI raised $350 million at a $3.5 billion valuation, nearly triple its price 17 months ago.
- Its annualised revenue jumped from roughly $20 million to more than $350 million in a year.
- The round underlines a broader shift: AI labs now pay for finished, expert-built data.
Snorkel AI just got a lot more expensive, and it’s not because it labels data faster than anyone else anymore. The company, which used to sell software for tagging training data, now builds and sells the finished datasets and simulated work environments that frontier labs train their models on.
That pivot has taken its annualised revenue from around $20 million a year ago to more than $350 million now, and investors have responded by tripling its valuation to $3.5 billion in a $350 million round.
Insight Partners and S32 co-led the raise, with existing backers Addition, Greylock, Wells Fargo, Lightspeed, GV and several others returning, alongside new investors including March Capital and Third Point Ventures. It brings Snorkel to nearly three times the $1.3 billion valuation it held after its $100 million round in May 2025.
From labelling software to a data factory
Snorkel was founded in 2019 by Alex Ratner, Chris Ré, Paroma Varma, Braden Hancock and Henry Ehrenberg, spinning out of Stanford’s AI Lab, where Ratner and Ré had built the original open-source Snorkel project. For years, the company sold tools that let enterprises label their own data.
Since launching a data-as-a-service business in September 2025, it has flipped that model: human specialists across coding, law and medicine now design tasks and grading rubrics, while a layer of AI models automates quality control, producing datasets and reinforcement-learning environments it sells directly to labs, hyperscalers, enterprises and the US federal government.
“Our strong view is that 100% of the data that labs will get value out of will have some human input in the foreseeable future. But 100% of that data will have to use synthetic and automated approaches to keep up with this complexity ,” said Ratner.
The data race is getting more competitive
Snorkel isn’t the only one cashing in on labs running out of usable internet text. Scale AI raised $1 billion at close to a $14 billion valuation before Meta later bought a 49% stake worth roughly $14.3 billion. Mercor went from a $30 million seed round to a $10 billion valuation on a similar bet, and has since been in talks to double that again. Surge AI, meanwhile, was reportedly weighing a first-ever outside raise at north of $25 billion.
Snorkel’s pitch against that crowd is that it designs the tasks and grades the output, which is a harder thing to compete away on price.
Snorkel says the capital will fund more researchers and engineers, expand its enterprise and government business, support third-party model evaluations, and push into new industries and data formats. It expects to turn a profit this year.
“Data is becoming more rare, more specialized, more difficult to find. If you want to train the most frontier, complex and capable models, now you need superior data,” said Andy Harrison, partner at S32 who co-led the funding.
If the last 18 months were about labs discovering they’d run out of internet to scrape, the next stretch is about who gets paid to manufacture what’s missing. Snorkel just put $3.5 billion that it can be the one doing the manufacturing — the harder question is whether that job stays defensible once every well-funded rival is building the same expert-plus-AI pitch.
