The new valuation is a sharp step up from May 2025, when Snorkel raised $100 million at a $1.3 billion valuation.
Much of that growth has followed a change in how Snorkel sells its technology. The company initially offered software that customers could use internally to automate data labeling. Last year, it shifted toward providing finished datasets and reinforcement-learning environments directly to customers through a data-as-a-service model.
Snorkel now works with frontier AI labs, hyperscalers, enterprises and the U.S. federal government. Coding has become one of its largest areas of demand.
CEO Alex Ratner founded the company following four years of research at a Stanford AI lab. Snorkel launched commercially in 2019 and has since expanded from its original labeling tools into systems designed to produce and evaluate more complex training material.
The company’s approach combines human expertise with its own software, models and AI agents. Specialists create scenarios, tasks and grading criteria, while automated systems handle portions of the production and quality-control process.
Ratner argues that increasingly sophisticated AI systems will continue to require both elements.
“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,” Ratner said. “But 100% of that data will have to use synthetic and automated approaches to keep up with this complexity.”
Snorkel draws from a network of tens of thousands of specialists in areas including coding, medicine and law. Rather than charging customers directly for those experts’ time, the company sells the finished datasets and environments they help produce.
That structure distinguishes Snorkel from businesses that operate primarily as marketplaces connecting AI companies with human specialists.
The distinction also matters when comparing revenue figures across the sector. Mercor has reached $2 billion in gross annualized revenue, while Handshake has crossed $1 billion and Micro1 has grown to $500 million. Those companies pay a substantial portion of their revenue to the specialists performing the underlying work.
Snorkel instead treats expert payments as part of the cost required to produce its datasets and RL environments. Its customers are buying the finished data product rather than purchasing human labor by the hour.
The company is betting that combining automation with expert involvement can make that production model more efficient as training requirements become increasingly specialized.
“Data is becoming more rare, more specialized, more difficult to find,” S32 partner Andy Harrison said. “If you want to train the most frontier, complex and capable models, now you need superior data.”
The new capital will support hiring across research and engineering as Snorkel expands its enterprise and government businesses. The company also plans to move further into third-party model evaluations, additional industries and new forms of training data.
Snorkel said it expects to become profitable this year even as growth remains its priority.
Its expansion reflects how the requirements for training AI models are changing. Developers increasingly need datasets designed around particular technical or professional tasks, along with environments where models can practice actions and receive feedback on the results.
Snorkel’s latest funding round gives it substantially more capital to pursue that demand, while its $3.5 billion valuation places a larger expectation on the company’s relatively new data-as-a-service business.
The next test will be whether the rapid growth behind its $375 million run-rate continues as AI developers decide how much specialized data production to outsource and how much to keep within their own organizations.
This analysis is based on reporting from The Star.
Image courtesy of Snorkel AI.
This article was generated with AI assistance and reviewed for accuracy and quality.
