Snorkel AI raises $350M at $3.5B valuation for training data

Snorkel AI has closed a $350 million Series E funding round at a $3.5 billion valuation, the company announced today. Insight and S32 led the investment, with participation from Alphabet's GV and more than half a dozen other backers.

The financing caps a remarkable year for the training data provider, which claims 18x growth and an annualized revenue run rate of $375 million since launching its current business model nearly twelve months ago.

From software platform to data service

Founded in 2019 by Stanford AI Lab researchers, Snorkel AI initially built software that automated the creation of labeled training datasets for supervised learning. That approach trains neural networks using datasets containing prompts paired with correct, human-generated responses.

The company's original product, Snorkel Flow, used statistical methods developed at Stanford to automate dataset labeling—a notoriously time-intensive process that had resisted earlier automation attempts due to accuracy problems.

Last year, Snorkel AI executed a fundamental business model shift. Rather than selling software tools for creating training data, the company now provides ready-to-use training datasets directly to customers. The pivot also expanded Snorkel's scope beyond supervised learning to include reinforcement learning, a more complex training methodology.

Reinforcement learning infrastructure at scale

Reinforcement learning differs from supervised learning in that datasets contain unanswered questions rather than prompt-response pairs. AI models must generate answers independently, which are then evaluated by human reviewers or automated systems that provide feedback to refine the model's reasoning.

Snorkel AI maintains a network of tens of thousands of human experts who generate reinforcement learning tasks. The company also provides the full technical infrastructure needed for training runs, including evaluation rubrics and specialized virtual environments.

These evaluation criteria can be extensive—spanning multiple pages for complex tasks like code generation, where reviewers must assess cybersecurity compliance, performance standards, and other technical requirements. Snorkel develops these rubrics for customers and continuously refines them based on reviewer feedback, particularly when two evaluators score the same AI response differently—a signal of inconsistent criteria.

The company also provisions training sandboxes tailored to specific use cases, such as simulated developer workstations for code generation models.

Why it matters

Snorkel AI's explosive growth reflects the training data bottleneck facing organizations building AI systems. As companies move beyond generic foundation models to specialized applications, they need domain-specific training data and evaluation infrastructure that most don't have the expertise or scale to build internally. Snorkel's shift from tools to turnkey data services addresses this gap directly, positioning the company as critical infrastructure in the AI development stack. The $375 million revenue run rate—achieved in under a year—suggests strong enterprise demand for outsourced training data operations.

Investment and roadmap

Co-founder and CEO Alex Ratner said the new capital will fund engineering hiring, AI safety initiatives, and support for open-source model evaluation benchmarks.

The details were first reported by SiliconANGLE.