How we researched and ranked AI data companies
Last updated: November 9, 2026
Scope
We define an AI data company as a business that sells human-created or human-verified data, or the expert workforce and tooling to produce it, for training and evaluating AI models. That covers annotation, RLHF and preference data, expert demonstrations, evaluation benchmarks and RL environments. Pure synthetic-data generators and general dataset marketplaces are outside this list.
Sources
- Official company sites, read on November 9, 2026: homepages, product pages, pricing pages and press releases.
- First-party funding announcements, such as Snorkel AI's Series E, Labelbox's Series E, Invisible's 2025 raise, Turing's Series E, micro1's Series A, Prolific's 2023 Series A and Sama's 2021 Series B.
- Independent reporting from outlets including Forbes, TIME, the Associated Press, Axios, TechCrunch and CNBC.
- Third-party compilations, such as a public AI-training-industry funding dataset and Sacra's company estimates, used only where first-party data was missing and always labelled.
Ranking criteria (in order of weight)
- Evidence of quality and frontier relevance: named customers in independent reporting, published benchmarks and research, and work on hard expert and agentic data.
- Breadth of offering: data types, modalities, RL environments, evaluation and tooling.
- Scale and staying power: disclosed funding, reported revenue, expert-network size and track record.
- Buyer considerations: ownership and neutrality, pricing transparency, geographic coverage and security posture.
- 2025–2026 momentum: new products, financings and acquisitions.
The final order is an editorial judgment that weighs these criteria together. It is not a mechanical numeric score, and it is not measured market share.
How we treat numbers
- Revenue, customer counts and network sizes from vendors are labelled company-reported.
- Valuations from deals still in talks are labelled reported or unconfirmed, never presented as completed rounds.
- Contract ceilings, such as Scale AI's Pentagon agreement, are described as ceilings, not revenue.
- Customer relationships are dated, because lab contracts change often.
- Funding entries show the latest round we could verify publicly. Where none was found, we say not disclosed.
What we did not do
- We did not test any product or service or commission paid pilot projects.
- We did not interview vendors. No vendor reviewed or approved this list.
- We did not publish user reviews or ratings.
Known limitations of this edition
- Turing's website blocked our automated reader, so its details come from cited secondary retrieval of its own pages and from Sacra.
- Surge AI's homepage contains little product detail, so we relied on its catalog pages and Forbes reporting.
- The Labelbox and Toloka pricing URLs we tried returned 404 errors. We report their pricing as not public.
- For Prolific and Sama, we found no financing announced after 2023 and 2021 respectively; unannounced private transactions may exist.
- Most private-company financials are unaudited.
Updates
The date at the top of each page shows when the evidence was last reviewed. Rankings may change as companies raise money, launch products or lose customers.