Snorkel AI raises $350M Series E at $3.5B valuation as training data demand surges

Snorkel AI, a startup that builds training datasets and simulated environments for AI labs and enterprises, has raised $350 million in a Series E round valuing it at $3.5 billion, TechCrunch reports. Insight Partners and S32 led the financing, with returning backers Addition, Lightspeed, Greylock, GV and Wells Fargo also chipping in.

The valuation nearly tripled in 17 months

According to TechCrunch, the new valuation is close to triple the $1.3 billion Snorkel commanded when it raised a $100 million Series D roughly a year and a half ago. The company went to market in 2019 after four years of research by co-founder and CEO Alex Ratner and his team at a Stanford AI lab, and TechCrunch describes it as a seven-year-old startup.

From labeling software to data-as-a-service

Snorkel started out selling software that automated the labeling of training data. As TechCrunch explains, it pivoted last year: instead of handing customers tools, it now delivers finished datasets, an offering it calls data-as-a-service. The company does not operate purely as a marketplace for human specialists. It uses a hybrid model in which its own software and models generate synthetic data alongside contributions from subject-matter experts, and it also sells reinforcement learning environments.

The company told TechCrunch its annualized revenue run-rate now stands at $375 million, an eighteenfold increase over the past 12 months. That growth, the report notes, is being propelled by AI labs competing fiercely for high-end training data.

A crowded market with fuzzy revenue math

Snorkel is not the only company riding this wave. TechCrunch points to several rivals positioning themselves as AI data labs that have scaled even faster on paper: Mercor's gross annualized revenue has reached $2 billion, Handshake passed the $1 billion mark earlier this year, and Micro1, per TechCrunch's own reporting, has grown to $500 million.

Those headline figures come with an important caveat. Firms in this space typically pay out roughly 60 to 70 percent of their top-line income to the domain specialists doing the work, which means their net annual revenue is far below the gross numbers they advertise.

Snorkel says its accounting works differently. Because it sells complete datasets and reinforcement learning environments rather than billable human labor, payments to its human experts are recorded in cost of goods sold rather than being subtracted from the annualized revenue it publicizes, according to the company.

Why it matters

A near-tripling of valuation in 17 months is a clear signal about where investors think the real bottleneck in AI now lies. Compute and model architecture have absorbed most of the industry's capital and attention, but as labs exhaust easily scraped web data, access to high-quality, expert-annotated, synthetic and simulated training data has become a scarce resource that companies will pay serious money to secure.

The round also shows how quickly data work is consolidating into dedicated, well-capitalized companies rather than remaining an ad hoc part of model development. And the gross-versus-net revenue distinction matters for anyone tracking the sector: with Snorkel, Mercor, Handshake and Micro1 all touting large run-rate figures, what each number actually measures differs from company to company, and conclusions about which business is bigger, or healthier, depend on reading the fine print.