Snorkel AI says the market for training data has moved beyond mass labeling. On Sept. 22, 2026, the company announced a $350 million Series E at a $3.5 billion valuation, co-led by Insight Partners and S32, and said the money will expand its work on data and environments for frontier AI systems.
The core claim is that the scarce commodity in AI has changed. In Snorkel’s framing, the older model of “Data 1.0” was mostly a volume problem: hire enough people, label enough examples, and scale output with headcount. The newer layer, which the company calls “Data 2.0,” is a research problem. It requires expert tasks, simulated environments and evaluation rubrics that can take qualified humans hours or days to design, because frontier systems need data that captures nuanced failure modes, reward signals and cheating behavior rather than just more labels.
That matters because it changes what customers are buying. If the bottleneck is no longer simple annotation, then the vendor with the most attractive economics is not necessarily the one that can supply the cheapest labor. It is the one that can package subject-matter expertise, software and model assistance into repeatable datasets, benchmarks and environments. Snorkel’s announcement says its investment will expand an “agentic data factory” and extend the company into vertical and enterprise AI, while also supporting open research through programs such as its Open Benchmarks Grants initiative.
Snorkel’s own blog post, published the same day, makes the company’s growth story more explicit. In that post, CEO Alex Ratner said the business had grown more than 18x since launching its data-as-a-service offering nearly a year ago and had crossed a $375 million annualized revenue run rate. Those numbers are important, but they are still company-reported claims in the material provided here; there is no independent audit, filing or third-party verification in the sources supplied for this article. Readers should treat them as management statements, not independently confirmed market data. Snorkel’s blog also lays out the argument that humans and AI agents now need to collaborate on data development because the frontier has become too complex for human-only workflows.
TechCrunch’s coverage adds a useful accounting distinction. It reported that Snorkel originally sold data-labeling automation and later shifted toward completed datasets and simulated environments, a model the company calls data-as-a-service. The publication also noted that Snorkel’s revenue framing is different from labor-marketplace businesses because payments to experts are treated as cost of goods sold rather than as gross revenue. That distinction matters for comparison: a company that sells finished data products can show a much cleaner top-line number than one that passes most of its intake straight through to contractors.
For AI labs, enterprise buyers and investors, the practical consequence is straightforward. As models get better at routine tasks, value moves toward harder-to-design environments that reveal whether a system can actually perform in realistic conditions. In coding, for example, that could mean tasks that approximate software work over days rather than single-turn prompts; in regulated sectors, it could mean evaluation sets that catch subtle errors before deployment. Snorkel is positioning itself as infrastructure for that shift, not just as a staffing layer.
The trade-off is that this kind of data is harder to standardize and probably more expensive to produce than commodity labeling. It may also be less transparent to outsiders, because much of the output depends on specialized expertise, bespoke environments and internal quality-control processes. That makes the funding round impressive as a signal of demand, but not a proof of durable margins or broad market adoption. The round and the revenue claim are meaningful because they point in the same direction; neither one, by itself, proves the economics are settled.
Snorkel says it will use the new capital to increase capacity, invest more in vertical and enterprise AI, and push its research into new domains and modalities. Watch whether Snorkel’s next disclosures specify audited revenue, customer mix or product detail, because those are the signals that separate durable demand from a one-time fundraising narrative.
