Snorkel AI raises $350M Series E at $3.5B valuation

Snorkel AI published that it raised $350 million in Series E financing at a $3.5 billion valuation, led by Insight Partners and S32, as demand from frontier labs for complex training data and reinforcement-learning environments accelerates.

FINANCE desk — the data layer feeding frontier labs just re-rated hard: $350M fresh capital at nearly 3× last year’s valuation.

Who led, as each page writes it. The blog says the round was led by Insight and S32, with participation from Third Point, March, Blumberg, Allegis, Standard VC, and Frontline, plus existing investors Addition, Lightspeed, Greylock, GV, P7, Wells Fargo, Walden Catalyst Ventures, and Factory. Reuters separately reported that Insight Partners and S32 led, and that existing investors Addition, Greylock, and Wells Fargo were among the participants. Reuters does not print the longer list. File the long list as the blog’s. File the shorter list as Reuters’. This desk did not see the shareholder agreement.

The revenue figures do not match, so both stay. The blog says Snorkel crossed an annualized revenue run rate of $375 million this week. Annualized revenue run rate, often shortened to ARR, takes a recent stretch of sales and stretches it to a full year. It is not a completed audit of the year. Reuters cited about $350 million in that run rate, up from roughly $20 million a year earlier, driven by a data-as-a-service business the wire says launched in September 2025. Data-as-a-service, here, means selling finished training data, not only software. Do not average $375 million and about $350 million into a third number. The blog also says the company has grown more than 18 times since that data offering launched nearly a year ago. More than 18 times is the blog’s growth claim. It is not Reuters’ “roughly $20 million” line rewritten.

The prior round is Reuters’. The wire says the new $3.5 billion valuation is nearly three times the $1.3 billion valuation from the last raise, $100 million in May 2025. Nearly three times is Reuters’ phrase for those two prices. This desk did not recompute a private share price. The May 2025 round is the comparison Reuters printed. It is not this filing’s event.

What Snorkel says it builds. The blog describes an agentic data platform: human experts working with specialized AI models and agents to make datasets and reinforcement-learning environments. Reinforcement learning, or RL, is a way of training a model by letting it try a task and scoring the attempt. An environment is the simulated task it practices in. Agentic, here, means software agents do part of the work alongside people. Reuters uses the phrase “agentic data development platform” and says it pairs human experts with specialized AI models and agents. The blog says sometimes hundreds of those agents per task type. Reuters says thousands of specialized models and agents. Hundreds per task and thousands overall are different sentences from different pages. File both. Do not merge them into one headcount. This desk did not use the product.

Who they say they serve. The blog says Snorkel now partners with frontier labs, large cloud companies (the page spells that word “hyperscalars”), neolabs, vertical AI leaders, enterprises, and U.S. government agencies. Reuters says frontier AI labs, hyperscalers, enterprises, and the U.S. federal government. A frontier lab is a company training the most advanced models. A hyperscaler is a giant cloud company. A vertical AI company, here, is one built for a single industry. Neolabs is the blog’s word. The page does not define it, so this desk does not invent a definition. File both customer lists as attributed. This desk did not see the contracts.

What the money is for, as Reuters reports it. Hire researchers and engineers. Expand enterprise and government work. Support third-party checks of other companies’ models. Push into new industry verticals and new kinds of data. A vertical is an industry. A modality is a kind of data, such as text, code, or images. The company said it expects to reach profitability this year. Expects is a plan. It is not a finished profit statement. File the spending lines and the profitability line as Reuters’. This desk did not see a hiring plan or the books.

Plain English for the rest of the card: Series E = a late private funding round. valuation = the price the round puts on the whole company. $3.5 billion is that price. ARR = recent revenue stretched to a year. The blog says $375 million. Reuters says about $350 million. data-as-a-service = selling finished training data. RL environment = a simulated task a model practices, with a score at the end. agentic = people plus software agents. hyperscaler = a giant cloud company. nearly three times = Reuters’ comparison of $3.5 billion with the May 2025 valuation of $1.3 billion. This filing is the 22 Sep announcement. It is not a public stock listing.

PRIMARY here: Snorkel’s 22 Sep 2026 blog, “Data 2.0 and the research era of AI data,” by Alex Ratner — Tier A PRIMARY, the company’s own record. Reuters’ same-day exclusive is an independent wire, not a substitute primary. The $350 million Series E, the $3.5 billion valuation, the blog’s investor list, the $375 million run rate, and the more-than-18-times growth line are the blog’s. The Insight Partners wording, the shorter participant list, the about-$350 million run rate, the roughly $20 million year-earlier line, the September 2025 data-as-a-service launch, the May 2025 $100 million round at $1.3 billion, the nearly-three-times line, the agentic data development platform phrase, the customer list Reuters prints, the use of capital, and the expectation of profitability this year are Reuters’. NOT claimed: a single ARR figure this desk averaged, an audited revenue number, that profitability is already booked, a cap table this desk saw, a customer contract this desk read, a stock tip, or investment advice. Distinct from the already-filed verda-189m, softbank-openai-bonds, and amd-1-trillion-market-cap.