Snorkel AI raised $350 million at a $3.5 billion valuation. More strikingly, chief executive Alex Ratner says annualized revenue passed $350 million, up from about $20 million a year ago, after the company pivoted in September 2025 from selling labeling software to delivering finished datasets and reinforcement-learning environments. Reuters, September 22
The change in product matters more than the valuation. Snorkel’s specialists in coding, law and medicine write scenarios, tasks and grading rubrics; thousands of models and agents perform much of the production and quality-control work. The buyer receives a training asset rather than seats in a tool. That turns expert judgment into inventory and lets the supplier price the output rather than the labor hours.
Ratner’s claim that every valuable frontier dataset will retain some human input is plausible but convenient for the seller. The important test is whether the resulting environments predict performance on genuinely novel tasks, rather than merely supporting another cycle of benchmark-shaped optimization. Snorkel expects profitability this year; no customer concentration or gross-margin figures were disclosed.