Snorkel AI Hits $3.5B as the Training Data Rush Heats Up

Opportunity assessment: high. The market for AI training data isn’t cooling. It’s speeding up, and investors are paying a lot to get in.

Snorkel AI has raised a $350 million Series E at a $3.5 billion valuation, according to TechCrunch AI. Insight Partners and S32 led the round. The valuation is nearly triple the $1.3 billion Snorkel got in its $100 million Series D just 17 months ago. Existing backers Addition, Lightspeed, Greylock, GV and Wells Fargo also put in more money.

📍 Situation Report

The key facts:

1. Funding: $350M Series E at a $3.5B valuation.
2. Revenue: Snorkel says its annualized run rate is now $375 million. That’s up eighteenfold in the last 12 months.
3. Age: The company is seven years old. It launched commercially in 2019 after four years of research by co-founder and CEO Alex Ratner and his team at a Stanford AI lab.
4. The pivot: Last year, Snorkel moved from selling data-labeling automation software to selling finished datasets. It calls this “data-as-a-service.”

That pivot explains most of the growth. Selling software tools puts Snorkel in the tooling budget. Selling finished datasets and reinforcement learning (RL) environments puts it in the budget AI labs spend on making their models better. The second budget is much bigger right now.

⚙️ How Snorkel Actually Works

Snorkel isn’t just a marketplace of human experts. It uses a hybrid model:

• Synthetic generation: Its own software and models produce training data automatically.
• Subject matter experts: Human specialists add the judgment and edge cases that synthetic data misses.
• RL environments: These are simulated settings where models practice tasks and get feedback. They’ve become core infrastructure for training reasoning and agentic models.

What stands out to me is that the hybrid approach is a bet on margins. If a company relies only on human experts, its costs grow with every hour of work. If software does more of the heavy lifting, each new dollar of revenue should cost less to deliver.

📊 The Competitive Field

Snorkel isn’t growing alone. TechCrunch AI lists several other data companies that call themselves AI data labs and are growing fast:

1. Mercor: gross annualized revenue of $2 billion.
2. Handshake: passed $1 billion earlier this year.
3. Micro1: reached $500 million.

There’s a catch in those numbers. These marketplaces pass roughly 60% to 70% of their top-line income straight to the domain specialists doing the work. So their net revenue is much lower than the gross figures in the headlines.

Snorkel is drawing a line here. The company says it sells RL environments and complete datasets, not human labor. So it books payments to its experts as cost of goods sold, and it argues that its $375 million figure isn’t built the same way as the gross marketplace numbers. That’s the company’s own framing. Still, it’s a real accounting difference to keep in mind when you compare these firms side by side.

🎯 Why This Matters

For years, the AI race was mostly about compute and model design. Now data is the bottleneck. The public internet has mostly been used up for pretraining. The next gains come from specialized, expert-level data and training environments built for reasoning and agent tasks.

This matters to practitioners in three ways:

1. Data is now bought, not scraped. Frontier labs are spending heavily on curated, domain-specific datasets. Expect enterprises building their own models to follow.
2. RL environments are becoming a product category. If you’re building agents, simulated training environments are turning into something you can buy, not only build yourself.
3. Expert work is getting paid well. Domain specialists like doctors, lawyers and engineers are becoming a key input in the AI supply chain.

🔭 What Comes Next

With $350 million in fresh capital, Snorkel will likely expand its RL environment offerings and push harder into enterprise accounts. The bigger question is whether the hybrid synthetic-plus-expert model can beat the pure human marketplaces on margins as the market matures.

Also watch how these companies report revenue. As more AI data labs raise at high valuations, investors will want to know the difference between gross and net revenue. Full details are available in the original TechCrunch AI report.