Snorkel AI Raises $350M Series E at $3.5B Valuation to Scale Data Labeling

Snorkel AI, a seven-year-old company that helps enterprises build and label training data for AI models, has closed a $350 million Series E round that values the startup at $3.5 billion. The funding, confirmed by TechCrunch, more than triples the company's previous valuation and marks one of the largest recent bets on the data infrastructure layer of the AI stack.

The Palo Alto-based company sells what it calls data-as-a-service: software that combines programmatic labeling, human review and model-assisted workflows to turn raw, unlabeled data into training sets. That approach targets a persistent bottleneck for companies deploying large language models and other AI systems — most organizations have plenty of raw text, images and logs, but lack the clean, annotated examples needed to fine-tune models for specific tasks.

Demand for that kind of tooling has surged alongside the generative AI boom. Foundation model developers and enterprise AI teams are spending heavily on data curation, labeling and evaluation, and Snorkel competes with a mix of legacy annotation vendors, open-source labeling tools and in-house data operations teams. The new capital gives Snorkel room to expand its sales motion, invest in product development and pursue acquisitions or partnerships as the market consolidates.

The round also reflects a broader investor appetite for picks-and-shovels AI companies — those selling infrastructure and services rather than end-user applications. While much of the attention and capital has gone to model builders like OpenAI, Anthropic and Google, the companies supplying the data, compute and tooling underneath them have attracted growing check sizes. Snorkel's jump from a roughly $1 billion valuation to $3.5 billion in a single round puts it in the upper tier of private AI infrastructure startups.

Snorkel has not disclosed its revenue or customer count, and the company declined to comment on specific financial metrics. Its challenge now is to convert a hot fundraising environment into durable enterprise contracts, especially as customers weigh build-versus-buy decisions and some AI budgets shift toward inference and deployment costs. The next signal will be whether Snorkel can show that its programmatic labeling approach scales across the messy, domain-specific data sets that large organizations actually hold.

For the wider AI industry, the deal reinforces a simple reality: model quality depends as much on data pipelines as on architecture and compute. If Snorkel can prove that thesis at enterprise scale, it will strengthen the case for a standalone data layer in the AI stack — and likely invite more competition from both startups and the cloud providers that already sell annotation services.