Behind the investment: Snorkel AI

Data has become one of the most important ingredients in advancing the frontier of AI and unlocking continuously improving capability, safety, and reliability.  As AI systems take on longer-running, higher-stakes work, the data needs from frontier AI labs and enterprises are shifting from generalist data that humans can write down to specialized data, environments, and evaluations that often need to be built. 

Delivering this next wave of “Data 2.0” requires the ability to simulate real-world environments that stay consistent and correct across multi-step agentic tasks in specialized domains. That, in turn, requires an engineering- and research-first approach to data collection, serving, evaluating, and scaling. 

Snorkel AI’s agentic data factory pairs AI tooling with a managed expert network to build the complex training data and reinforcement learning (RL) environments needed for Data 2.0. Its customers include leading frontier AI labs, emerging research labs, fast-growing AI-native applications, and large enterprises. Today, we are excited to co-lead Snorkel AI’s $350 Series E alongside our friends at S32, with participation from Third Point, March, Blumberg, Allegis, Standard VC, Frontline, and existing investors.  

A shifting tide of data needs for frontier AI

The past few years have brought remarkable results across AI model training and the real-world applications built on top of it. Even as models get smarter and compute demands rise, the supply of high-quality, specialized data is limited, often difficult to collect or produce, and requires specialized expertise to structure, validate, and deliver. 

For years, the defining challenge in AI training data was volume. This was the era of “Data 1.0”: simple, well-defined labeling tasks (check an image, flag a response, rank two outputs), which was well-suited for large contractor pools. 

But AI labs, AI-powered applications, and enterprises are now facing problems that are structurally different. Demand is shifting toward expert agentic tasks, environments, and evaluation rubrics that take skilled humans and machines working together to construct. Reliable and capable real-world agentic systems need the simulated environments where models can practice and learn on the hardest real-world tasks, and the evaluation systems that prove they got better.  

These environments can’t be easily crowdsourced. They need to be engineered: tool-equipped and grounded in real operating contexts so that agents can attempt a long, multi-step task and be evaluated fairly on the output. The environment itself becomes part of the data, and building these environments and comprehensive datasets requires an increasingly tech-first approach, along with research-level skills and domain expertise. 

This is what we mean by Data 2.0: data that cannot simply be written down but that needs to be built. Data 2.0 is becoming an increasingly critical layer of the AI supply chain. 

Enter: Snorkel AI

We have been watching the AI data space for years with a mixture of excitement and skepticism. The excitement stems from data’s obvious strategic importance to AI labs and enterprises across model training, fine-tuning, and evaluation. The skepticism comes from questions we kept asking about even the most successful businesses in this space: Could they build durable platforms? How would pricing evolve as the market matured?  

Snorkel AI is a company we’ve followed since its early days, when Alex, Paroma, Chris, and the founding team took their deep research around data-centric AI at the Stanford AI Lab and began applying it to the world of enterprise data. We’ve consistently heard great things about the Snorkel team, the quality of their product, and their depth of expertise in data-centric AI. 

Over time, we got increasingly excited by the team’s depth of technical and research expertise combined with the clear market pull they are seeing from customers and partners.   

Snorkel’s agentic data factory pairs expert human judgment with AI tooling in a tight feedback loop, unlocking data production at a speed and level of complexity that neither a human-only nor tech-only approach could achieve alone, with automated quality checks running at every step before anything ships. 

Partnering with Snorkel AI

Snorkel’s team has a strong blend of academic research at the forefront of data evaluation and AI training, deep product and technological clarity, and both AI and application-level knowledge. Snorkel helps fill the gaps where frontier models hunger for more data today (specialized domains, hard-to-define tasks, and benchmark blind spots) to improve both trust and capability.  

We are looking forward to all that Alex, Paroma, Vincent, Henry, Chris, and the team continue to build and execute. 

Snorkel AI is an Insight Partners portfolio company.