Snorkel AI funding: Every round, investor, and valuation (2026)

Snorkel AI makes the data that AI models learn from. It started as software that let companies label training data with code instead of by hand, and it now runs what it calls an "agentic data factory": a network of human experts and AI agents that builds datasets, evaluations, and reinforcement learning environments for frontier labs, hyperscalers, banks, and US government agencies.

Snorkel AI has raised about $585 million across six rounds since it spun out of the Stanford AI Lab in 2019, all of it equity. Its latest, a $350 million Series E co-led by Insight Partners and S32 in September 2026, valued the company at $3.5 billion.

The investor list mixes the venture firms that backed the research project with the financial companies that became its first big customers. Greylock, Lightspeed, and GV sit alongside BlackRock, BNY, Wells Fargo, QBE, and Accenture. Let's dive into Snorkel AI's funding history.

Snorkel AI's funding comes in two bursts. It raised $135 million in its first two years as an enterprise software company, went almost four years without a priced round, and then raised $450 million in 16 months once it started selling data to AI labs.

Snorkel began as an open-source research project at the Stanford AI Lab. Its idea was weak supervision: instead of paying people to label thousands of examples one by one, a data scientist writes short rules, called labeling functions, and the software combines them into training labels. Google, Intel, and others used the open-source version before the company existed.

CEO Alex Ratner, Stanford professor Chris Ré, Braden Hancock, Paroma Varma, and Henry Ehrenberg founded the company in 2019. Greylock and Google's venture arm GV co-led the seed that spring, and Greylock also gave the team office space. Greylock led a Series A that fall, with GV and the CIA-backed fund In-Q-Tel alongside.

Snorkel kept both rounds quiet until it came out of stealth in July 2020, when it disclosed $15 million in combined funding without splitting the two. Greylock partner Saam Motamedi joined the board.

Snorkel raised twice in 2021, when investors were paying up for AI infrastructure. Lightspeed led the Series B in April, and partners Ravi Mhatre and Raviraj Jain joined the board. BlackRock came in for the first time.

Four months later, BlackRock and Addition, Lee Fixel's growth fund, co-led the Series C and made Snorkel a unicorn. The pitch was Snorkel Flow, a platform for "data-centric AI" that let banks, insurers, and government agencies build models on their own documents without sending them to outside labelers.

The long gap between the Series C and Series D was filled by customers. QBE, the Australian insurer, invested in January 2024 after its North American teams used Snorkel Flow for claims and underwriting. BNY signed a strategic agreement in 2024 and had become an investor by the following year.

Addition returned to lead the Series D in May 2025, and Prosperity7, the growth fund of Saudi Aramco's venture arm, joined. The round launched Snorkel Evaluate, a tool for testing specialized models, and an expert data service for AI labs. Anthropic's Kate Jensen gave a supporting quote in the announcement, an early sign of which customers the new service was built for. Accenture Ventures invested three months later and brought Snorkel into its financial services work.

The Series E was priced on a different business. In September 2025, Snorkel launched data-as-a-service, selling finished datasets and reinforcement learning environments built by its expert network rather than software licenses. The company says that line grew more than 18 times in its first year.

Insight Partners, through managing director Lonne Jaffe, and S32, the firm run by former Google executive Andy Harrison, co-led the round. Third Point Ventures, March Capital, and four other firms were new, and Wells Fargo appeared on the list as an existing investor.

The customer base moved with the product. Snorkel now describes itself as a frontier AI data lab, and Reuters reported that it had shifted from selling software to supplying finished datasets and reinforcement learning environments, the simulated tasks labs use to train models that write code or operate software. Coding data is one of the largest sources of demand.

Snorkel AI's valuation barely moved for four years and then nearly tripled. The flat stretch matches the period when it was an enterprise software company, and the jump matches the switch to selling data.

The Series D priced Snorkel at 1.3 times its Series C value after almost four years, a modest step during a period when AI startups reached unicorn status far faster than older software companies. The Series E then valued it at 2.7 times the Series D, only 16 months later.

Unlike most AI companies at this stage, Snorkel discloses a revenue figure to set against the price. Ratner wrote in September 2026 that the company's annual run rate had crossed $375 million. Reuters, citing the company, reported about $350 million, up from roughly $20 million a year earlier.

At $3.5 billion, Snorkel is valued at under ten times its run rate. That is a lower multiple than many AI application companies command, and it reflects how investors price data businesses: much of the revenue pays for expert labor, so margins sit below those of pure software. The company did not say whether $3.5 billion is a pre-money or post-money figure.

Snorkel AI's investors fall into three groups: the early venture backers that funded the Stanford research, growth funds that led the later rounds, and financial companies that invested after becoming customers.

Greylock is the founding investor. It co-led the seed, provided office space, led the Series A, and has joined every priced round since. Motamedi has held the firm's board seat since the company came out of stealth.

Lightspeed led the Series B and took two board seats, held by Mhatre and Jain. The firm says it also took part in the seed. It returned in the Series C, the Series D, and the Series E, which makes it one of the few investors in every round since 2021.

Addition is the investor most tied to Snorkel's later growth. It co-led the Series C, led the Series D, and took "significant participation" in the Series E. Partner Todd Arfman led the 2025 deal, when the company was still mostly a software business and the expert data service was new.

The two co-leads of the Series E are new to the company. Insight Partners wants Snorkel to take its data factory into healthcare, law, and software engineering, fields where training data has to come from licensed professionals. S32 led alongside it, and neither firm disclosed a board seat.

GV, the venture arm funded by Google's ad and cloud profits, has invested since the Series A and returned in the Series E. In-Q-Tel, the nonprofit fund that invests for the CIA and other US agencies, joined early as well. Government work is still part of the business: Snorkel holds a $1.24 million Air Force research contract and names US agencies among its customers.

The financial investors are Snorkel's most distinctive backers. BlackRock co-led the Series C. BNY, QBE, and Wells Fargo invested on the strategic side, and Ratner said in 2024 that most of the largest US banks were customers.

Accenture Ventures invested in August 2025 to build industry solutions with Snorkel, starting with financial services. The amounts behind these strategic deals have not been disclosed. The arrangement gives Snorkel reference customers in a regulated industry, and it means part of its revenue comes from its own shareholders.

Snorkel AI spends its funding on its expert network, its agent platform, research, and new markets. Reuters reported that the company expects to reach profitability in 2026, so the Series E is growth capital more than a cash runway.

The largest cost is people who are not on the payroll. Snorkel works with tens of thousands of specialists in coding, law, medicine, and more than 1,000 other domains, and it pays them per project. Its recruiting page says top experts earn more than $3,000 a week.

Ratner argues that Snorkel sells data products, not hours of labor. That lets it pay experts more than a staffing marketplace would while protecting its margin. It is the main difference from rivals such as Mercor, which raised its own $350 million Series C to build a marketplace that places experts with labs.

The software side of the business now checks the humans' work. Snorkel runs hundreds of specialized AI agents per task type to review and score expert submissions. The company says the system improves quality control efficiency by more than 50% on coding data, which is one of its biggest demand areas.

That is where the Series E is aimed. The press release lists four uses: expanding capacity of the data factory, investing in vertical and enterprise AI, extending research into new domains and data types, and funding open research.

Snorkel keeps a research arm, a legacy of its Stanford roots. Its Open Benchmarks Grants program has committed $3 million to public AI benchmarks such as Terminal-Bench and OSWorld 2.0, and the company plans to expand it. Benchmarks are a sales tool as much as research, much like LMArena's leaderboard business: the labs that use a test to measure their models need data to improve on it.

Snorkel cut 31 of its 240 employees in September 2025, about 13%, months after the Series D. Business Insider reported that software engineering took the largest share of cuts as the company "deprioritized" legacy areas to focus on data-as-a-service. The Series E calls for hiring researchers and engineers again.

Snorkel AI's growth push is aimed at the frontier labs and at specialist fields. The labs are flush with cash, as Anthropic's $65 billion Series H and OpenAI's record funding rounds show, and a growing share of their spending goes to expert data and reinforcement learning environments. Snorkel also plans to expand its enterprise and government business, run third-party model evaluations, and spend a large share of its research budget on safety and alignment data.

The competitive picture changed in June 2025, when Meta's 49% stake in Scale AI pushed labs such as Google to look for other suppliers. That opened the market for Snorkel, Surge AI, and Mercor, which now compete for the same experts and the same contracts.

The company has no public IPO plans, and its expected profitability reduces the need to raise again soon. The main risks are concentration and dependence: Snorkel does not name its lab customers, a handful of them likely account for much of its revenue, and those labs could build more of their data pipelines in-house.

Snorkel AI has raised about $585 million across six rounds, from a seed and Series A totaling $15 million in 2019 to a $350 million Series E in September 2026. QBE, BNY, Accenture, and Wells Fargo have also made strategic investments of undisclosed size. The company has not reported any debt.

Snorkel AI's lead investors are Greylock, Lightspeed, Addition, BlackRock, Insight Partners, and S32. Other backers include GV, In-Q-Tel, Prosperity7, Factory, Walden Catalyst, Third Point Ventures, March Capital, and financial companies including BNY, Wells Fargo, QBE, and Accenture.

Snorkel AI was valued at $3.5 billion in its September 2026 Series E. That is up from $1.3 billion in May 2025 and $1 billion in August 2021.

No. Snorkel AI is private and has not filed for an IPO. Its shares are not traded on any stock exchange.

Snorkel AI is owned by its founders, employees, and investors. Co-founder Alex Ratner is CEO. The largest outside holders are likely Greylock, Lightspeed, and Addition, which have each invested in several rounds, though exact stakes are not public.