Snorkel's $3.5bn round prices the company at about nine times its claimed run rate

The valuation and the revenue claim moved at different speeds. Snorkel's new price is nearly 2.7 times the $1.3 billion it carried in May 2025 ^[5][2]. Its run rate, on the company's own account, is more than 18 times the level around the launch of its expert-data service roughly a year earlier ^[9]. At $3.5 billion against $375 million, the Series E values the company at about 9.3 times the claimed run rate ^[1]. Run the same sum for the Series D, using the baseline implied by Snorkel's 18-fold claim, and the figure lands above 60 times ^[3][2]. The comparison is loose, because the $1.3 billion is dated May 2025 and the run-rate baseline sits around September 2025 ^[2][9].

So this is the cheaper of the two rounds per dollar of claimed revenue ^[1][3]. The $375 million is a company figure. Per implicator.ai, a run rate projects a recent period across a full year; it is not booked annual revenue or audited sales, and it is not evidence that the company is profitable ^[10][8].

Margin is where a data business of this shape is decided, and Snorkel's accounting makes that explicit: it sells the finished product and treats payments to its experts as a cost of producing it ^[13]. Other suppliers report gross marketplace revenue, which implicator.ai says is not directly comparable with Snorkel's number ^[13]. The same $375 million could sit on very different margins depending on what tens of thousands of specialists in fields including coding, law and medicine are paid ^[12]. The captured materials leave out gross margins, audited revenue, customer concentration and independent tests comparing Snorkel's datasets with alternatives ^[18].

The product description is specific about evaluation. The environments present models with tasks in simulated settings and supply criteria for grading their responses, with human specialists writing the scenarios and rubrics while Snorkel's software and specialized models generate material and check its quality ^[11]. A graded environment is not a labeled example set, and Snorkel's 2019 product was software that automated parts of labeling by letting specialists write rules ^[7][6]. The only guide the release offers to which line of business grew is the customer list, which runs to frontier AI labs, hyperscalers, enterprises and U.S. government agencies, with coding among the largest areas ^[14][15].

The launch date is unsettled in Snorkel's own materials. A post dated May 29, 2025 described Snorkel Evaluate and Expert Data-as-a-Service as generally available and already serving LLM developers ^[16]. The September 2026 release calls September 2025 the service's launch, and implicator.ai says the sources do not explain the difference ^[17]. Since the 18-fold growth claim is measured from around that launch, the baseline moves with whichever date holds ^[9].

On the demand side the independent evidence is indirect. An EDM Association survey published in May 2026, covering more than 435 organizations in more than 50 countries, found about 31% reporting advanced data-strategy capability, and implicator.ai notes the survey measures organizational readiness, not Snorkel's data quality or demand for its products ^[19]. Accenture invested in Snorkel and announced a financial-services collaboration in August 2025; that announcement gave no customer spending figures and no measured improvements from the work ^[20]. A team deciding this quarter whether to buy graded environments will be doing it on its own pilot results.