Training data provider Snorkel AI raises $350M at $3.5B valuation
SiliconANGLE Maria Deutscher ● Covered by 2 sources
Snorkel AI just raised $350M at a $3.5B valuation. It’s betting that training data, evals, and safety are now the real AI bottleneck.
Based on reporting by SiliconANGLE, Maria Deutscher — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Snorkel AI has raised $350 million in a Series E round led by Insight and S32, with Alphabet’s GV among the backers. The company says the new money will go toward hiring more engineers, plus AI safety work and open-source model evaluation benchmarks.
That’s a big show of confidence in a company that started in 2019 with a much narrower pitch. Snorkel was founded by Stanford AI Lab researchers and first sold Snorkel Flow, software meant to cut the pain out of supervised learning, where models are trained on labeled examples. Its founders used statistical methods from their Stanford work to automate part of that labeling process and, according to the company, avoid the accuracy problems that had tripped up earlier attempts.
Last year, though, Snorkel changed course. Instead of just helping customers build training data, it moved into providing ready-to-use datasets. It also pushed beyond supervised learning into reinforcement learning, a more demanding setup where models face unanswered questions and then get judged after they respond. Those tasks often need human reviewers, and Snorkel says it relies on tens of thousands of experts to create them.
The company now bundles more than datasets. It also supplies evaluation rubrics, improves them over time when reviewers disagree, and provides training sandboxes for models that need virtual environments. In practice, that means it is selling a lot of the unglamorous infrastructure around AI training. That’s the business now, and the latest funding round suggests investors are happy to pay for the plumbing.
My take — AI-written commentary, not fact-checked reporting
The market is finally admitting that AI is not just about bigger models and louder demos. The dull stuff — data, evals, safety, and the miserable paperwork around all three — is where the money starts to stick. Open-source benchmarks are nice, but the real moat is still whoever can make humans and models agree on what “good” looks like without starting a fire.
Read more about this at: SiliconANGLE