Enterprise AI training data providers and what they supply
6 of 12 enterprise AI training data providers whose own websites we read say they source data by licensing it from the companies and rights-holders that own it, read 8 October 2026.
How it says it sources the data
Main kind of data it supplies
Modalities it names
Workforce or network size it states
Runs a programme for data owners to supply data
Every answer from every provider
A provider per row. Every answer opens onto the line it comes from, quoted exactly, with the page it sits in and the day of the read. "Not stated" marks a question the page leaves unanswered.
| Provider | Main kind of data it supplies | Modalities it names | How it says it sources the data | Workforce or network size it states | Runs a programme for data owners to supply data |
|---|---|---|---|---|---|
| Appen | varies by document | annotation, labeling, and collection for machine learning across text, image, audio… | a crowd workforce | Appen's global network spans 170 countries, includes verified domain specialists across… Not addressed in Appen off-the-shelf dataset catalog | not statedNot addressed in Appen AI training data services page Not addressed in Appen off-the-shelf dataset catalog |
| Defined.ai | data marketplace | 99,500 hours of meeting recordings in English with video, audio, transcripts, metadata… | not statedNot addressed in Defined.ai dataset marketplace | not statedNot addressed in Defined.ai dataset marketplace | has a data partner programme |
| Innodata | not statedNot addressed in Innodata home page | not statedNot addressed in Innodata home page | vetted experts | not statedNot addressed in Innodata home page | not statedNot addressed in Innodata home page |
| Mercor | licensed corpora | not statedNot addressed in Mercor Data page | licensed rights-holders | not statedNot addressed in Mercor Data page | has a data partner programme |
| micro1 | licensed corpora | not statedNot addressed in micro1 company referral page | licensed rights-holders | not statedNot addressed in micro1 company referral page | has a data partner programme |
| Nyne | licensed corpora | File contents plus their metadata, authorship, timestamps, and access permissions. | licensed rights-holders | not statedNot addressed in Nyne data partnerships page | has a data partner programme |
| Protege | licensed corpora Not addressed in Protege for data providers page | The Series A extension follows rapid adoption across healthcare, media, audio, motion… | licensed rights-holders | has a data partner programme | |
| Scale AI | annotation and labelling | Supported Annotation Types Scale Text Document Processing Natural Language Processing… | vetted experts | not statedNot addressed in Scale Data Engine page | not statedNot addressed in Scale Data Engine page |
| SimpleClosure AssetHub | data marketplace | That can include connected information across tools like Slack, Google Drive, Email… | licensed rights-holders | not statedNot addressed in AssetHub product page | has a data partner programme |
| Snorkel AI | expert or RLHF data | Standard & custom environments Repo + CLI tools Browser/GUI harness Multi-step/stateful… | vetted experts | not statedNot addressed in Snorkel AI home page | not statedNot addressed in Snorkel AI home page |
| Surge AI | expert or RLHF data | not statedNot addressed in Surge AI home page | not statedNot addressed in Surge AI home page | not statedNot addressed in Surge AI home page | not statedNot addressed in Surge AI home page |
| Troveo | licensed corpora | Buyers pay a premium for work they can follow across communication channels, knowledge… | licensed rights-holders | not statedNot addressed in Troveo business data page | has a data partner programme |
The source sentence behind every answer (70)
- Appen, Appen AI training data services page, read 2026-10-08, Main kind of data it supplies:
The world's leading AI training data provider, annotation, labeling, and collection for machine learning across text, image, audio, video, and geospatial data.
- Appen, Appen AI training data services page, read 2026-10-08, Modalities it names:
annotation, labeling, and collection for machine learning across text, image, audio, video, and geospatial data
- Appen, Appen AI training data services page, read 2026-10-08, How it says it sources the data:
Appen's global network spans 170 countries, includes verified domain specialists across 50 fields
- Appen, Appen AI training data services page, read 2026-10-08, Workforce or network size it states:
Appen's global network spans 170 countries, includes verified domain specialists across 50 fields
- Appen, Appen AI training data services page, read 2026-10-08, Runs a programme for data owners to supply data: No answer on the page, which ran to 4,613 characters when read on 2026-10-08.
- Appen, Appen off-the-shelf dataset catalog, read 2026-10-08, Main kind of data it supplies:
Off-the-shelf (OTS) AI training datasets are pre-collected, pre-licensed datasets that teams can license and use immediately for model training
- Appen, Appen off-the-shelf dataset catalog, read 2026-10-08, Modalities it names:
597 datasets across nine categories, from reinforcement learning (RL) tasks and verifiers to speech, code, and enterprise data.
- Appen, Appen off-the-shelf dataset catalog, read 2026-10-08, How it says it sources the data:
Contributor consent and documented licensing. Perpetual, non-exclusive commercial training rights.
- Appen, Appen off-the-shelf dataset catalog, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 6,844 characters when read on 2026-10-08.
- Appen, Appen off-the-shelf dataset catalog, read 2026-10-08, Runs a programme for data owners to supply data: No answer on the page, which ran to 6,844 characters when read on 2026-10-08.
- Defined.ai, Defined.ai dataset marketplace, read 2026-10-08, Main kind of data it supplies:
Showing 10 of 819 datasets
- Defined.ai, Defined.ai dataset marketplace, read 2026-10-08, Modalities it names:
99,500 hours of meeting recordings in English with video, audio, transcripts, metadata, action items, and recaps across industries.
- Defined.ai, Defined.ai dataset marketplace, read 2026-10-08, How it says it sources the data: No answer on the page, which ran to 3,418 characters when read on 2026-10-08.
- Defined.ai, Defined.ai dataset marketplace, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 3,418 characters when read on 2026-10-08.
- Defined.ai, Defined.ai dataset marketplace, read 2026-10-08, Runs a programme for data owners to supply data:
Partnerships Data Partner AI Service Supplier
- Innodata, Innodata home page, read 2026-10-08, Main kind of data it supplies: No answer on the page, which ran to 1,593 characters when read on 2026-10-08.
- Innodata, Innodata home page, read 2026-10-08, Modalities it names: No answer on the page, which ran to 1,593 characters when read on 2026-10-08.
- Innodata, Innodata home page, read 2026-10-08, How it says it sources the data:
providing the data, evaluation frameworks, and human expertise required to build AI systems that can be trusted at scale
- Innodata, Innodata home page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 1,593 characters when read on 2026-10-08.
- Innodata, Innodata home page, read 2026-10-08, Runs a programme for data owners to supply data: No answer on the page, which ran to 1,593 characters when read on 2026-10-08.
- Mercor, Mercor Data page, read 2026-10-08, Main kind of data it supplies:
The data your business generates every day is exactly what frontier labs need. We license it, package it, and pay you for it.
- Mercor, Mercor Data page, read 2026-10-08, Modalities it names: No answer on the page, which ran to 4,081 characters when read on 2026-10-08.
- Mercor, Mercor Data page, read 2026-10-08, How it says it sources the data:
The data your business generates every day is exactly what frontier labs need. We license it, package it, and pay you for it.
- Mercor, Mercor Data page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 4,081 characters when read on 2026-10-08.
- Mercor, Mercor Data page, read 2026-10-08, Runs a programme for data owners to supply data:
The data your business generates every day is exactly what frontier labs need. We license it, package it, and pay you for it.
- micro1, micro1 company referral page, read 2026-10-08, Main kind of data it supplies:
Operational data partnership using existing business documentation and operational datasets.
- micro1, micro1 company referral page, read 2026-10-08, Modalities it names: No answer on the page, which ran to 4,561 characters when read on 2026-10-08.
- micro1, micro1 company referral page, read 2026-10-08, How it says it sources the data:
Operational data partnership using existing business documentation and operational datasets.
- micro1, micro1 company referral page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 4,561 characters when read on 2026-10-08.
- micro1, micro1 company referral page, read 2026-10-08, Runs a programme for data owners to supply data:
The company completes onboarding and is approved as an active partner in micro1's Data Partnerships program.
- Nyne, Nyne data partnerships page, read 2026-10-08, Main kind of data it supplies:
We partner with you to license your company's data.
- Nyne, Nyne data partnerships page, read 2026-10-08, Modalities it names:
File contents plus their metadata, authorship, timestamps, and access permissions.
- Nyne, Nyne data partnerships page, read 2026-10-08, How it says it sources the data:
We partner with you to license your company's data.
- Nyne, Nyne data partnerships page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 3,973 characters when read on 2026-10-08.
- Nyne, Nyne data partnerships page, read 2026-10-08, Runs a programme for data owners to supply data:
Nyne turns your de-identified operational workflows into frontier-grade training data, under your governance and on your terms.
- Protege, Protege home page, read 2026-10-08, Main kind of data it supplies:
Protege streamlines access to private and proprietary real-world data across domains.
- Protege, Protege home page, read 2026-10-08, Modalities it names:
The Series A extension follows rapid adoption across healthcare, media, audio, motion capture, and more
- Protege, Protege home page, read 2026-10-08, How it says it sources the data:
Monetize your existing data assets or content with Protege — while maintaining clear rights protections and provenance.
- Protege, Protege home page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 4,197 characters when read on 2026-10-08.
- Protege, Protege home page, read 2026-10-08, Runs a programme for data owners to supply data:
Monetize your existing data assets or content with Protege — while maintaining clear rights protections and provenance.
- Protege, Protege for data providers page, read 2026-10-08, Main kind of data it supplies: No answer on the page, which ran to 1,626 characters when read on 2026-10-08.
- Protege, Protege for data providers page, read 2026-10-08, Modalities it names:
Protege works with data and content providers across domains. Select an industry below to learn more.
- Protege, Protege for data providers page, read 2026-10-08, How it says it sources the data:
Unlock New Revenue Through Data Licensing for AI
- Protege, Protege for data providers page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 1,626 characters when read on 2026-10-08.
- Protege, Protege for data providers page, read 2026-10-08, Runs a programme for data owners to supply data:
Protege works with data and content providers across domains. Select an industry below to learn more.
- Scale AI, Scale Data Engine page, read 2026-10-08, Main kind of data it supplies:
Collect, Curate, and annotate data. Train models and evaluate. Repeat.
- Scale AI, Scale Data Engine page, read 2026-10-08, Modalities it names:
Supported Annotation Types Scale Text Document Processing Natural Language Processing Transcription Content & Language Scale Image
- Scale AI, Scale Data Engine page, read 2026-10-08, How it says it sources the data:
Scale can provide the core tenet of any dataset with high-quality labels from domain experts.
- Scale AI, Scale Data Engine page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 3,423 characters when read on 2026-10-08.
- Scale AI, Scale Data Engine page, read 2026-10-08, Runs a programme for data owners to supply data: No answer on the page, which ran to 3,423 characters when read on 2026-10-08.
- SimpleClosure AssetHub, AssetHub product page, read 2026-10-08, Main kind of data it supplies:
AssetHub connects qualifying assets with vetted buyers, including AI labs, model developers, and other companies looking to license or acquire real-world code.
- SimpleClosure AssetHub, AssetHub product page, read 2026-10-08, Modalities it names:
That can include connected information across tools like Slack, Google Drive, Email, Notion, Figma, and more
- SimpleClosure AssetHub, AssetHub product page, read 2026-10-08, How it says it sources the data:
AssetHub helps assess, package, and connect SourceCode and WorkspaceData with AI buyers so founders can generate value through sale or licensing opportunities.
- SimpleClosure AssetHub, AssetHub product page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 6,730 characters when read on 2026-10-08.
- SimpleClosure AssetHub, AssetHub product page, read 2026-10-08, Runs a programme for data owners to supply data:
AssetHub helps startups monetize their assets, like SourceCode or WorkspaceData to vetted AI buyers looking for real-world work product.
- Snorkel AI, Snorkel AI home page, read 2026-10-08, Main kind of data it supplies:
Snorkel helps frontier labs and AI teams develop specialized training data and environments that set their models and agents apart.
- Snorkel AI, Snorkel AI home page, read 2026-10-08, Modalities it names:
Standard & custom environments Repo + CLI tools Browser/GUI harness Multi-step/stateful workflows Simulated environments
- Snorkel AI, Snorkel AI home page, read 2026-10-08, How it says it sources the data:
Well-specified expert-level tasks
- Snorkel AI, Snorkel AI home page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 18,397 characters when read on 2026-10-08.
- Snorkel AI, Snorkel AI home page, read 2026-10-08, Runs a programme for data owners to supply data: No answer on the page, which ran to 18,397 characters when read on 2026-10-08.
- Surge AI, Surge AI home page, read 2026-10-08, Main kind of data it supplies:
New Frontier data and RL environments, off the shelf
- Surge AI, Surge AI home page, read 2026-10-08, Modalities it names: No answer on the page, which ran to 949 characters when read on 2026-10-08.
- Surge AI, Surge AI home page, read 2026-10-08, How it says it sources the data: No answer on the page, which ran to 949 characters when read on 2026-10-08.
- Surge AI, Surge AI home page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 949 characters when read on 2026-10-08.
- Surge AI, Surge AI home page, read 2026-10-08, Runs a programme for data owners to supply data: No answer on the page, which ran to 949 characters when read on 2026-10-08.
- Troveo, Troveo business data page, read 2026-10-08, Main kind of data it supplies:
AI labs want to learn from companies like yours. Troveo matches you with them, anonymizes your data to a documented standard, and licenses it on terms you approve
- Troveo, Troveo business data page, read 2026-10-08, Modalities it names:
Buyers pay a premium for work they can follow across communication channels, knowledge and code repositories, and operational tools.
- Troveo, Troveo business data page, read 2026-10-08, How it says it sources the data:
Thousands of companies trust Troveo to license their data, protect their rights, and pay them on time.
- Troveo, Troveo business data page, read 2026-10-08, Workforce or network size it states: No answer on the page, which ran to 4,973 characters when read on 2026-10-08.
- Troveo, Troveo business data page, read 2026-10-08, Runs a programme for data owners to supply data:
Thousands of companies trust Troveo to license their data, protect their rights, and pay them on time.
Set aside after reading
3 named documents were read and left out because they are not what this board covers; each reason is below. They are counted, never answered.
- scale.com: Agentic Solutions for Enterprise | Scale AI: Scale's enterprise page is about building and deploying agentic AI systems for companies, not about supplying training data
- openai.com: OpenAI Data Partnerships | OpenAI: OpenAI's data partnerships announcement is an AI lab asking for data, not a provider supplying it
- welocalize.com: Welocalize | Translation & Localization for Global Businesses: Welocalize's home page is about translation and localisation services; it says nothing about supplying AI training data
The method in full
Each provider's own own websites were read through estate-fetch and stored whole (sha256 on every row).
For every page, five questions were answered from its own words: main kind of data it supplies, modalities it names, how it says it sources the data, workforce or network size it states, runs a programme for data owners to supply data.
Every stated answer carries the sentence it rests on, checked against the stored text; 'not stated' means the page read does not address the question.
A provider with several pages reads 'varies' where they disagree.
12 providers, 14 pages.
2 named documents are still to be read and are not counted yet.
Source list (19)
- scale.com: Scale Data Engine | AI Training Data at Scale | Scale AI (from the publisher's own own websites; reproduced for comment)
- scale.com: Agentic Solutions for Enterprise | Scale AI (read, set aside)
- surgehq.ai: Surge AI | Human Intelligence for AGI (from the publisher's own own websites; reproduced for comment)
- appen.com: AI Training Data Services | Appen (from the publisher's own own websites; reproduced for comment)
- appen.com: AI training datasets: off-the-shelf data catalog | Appen (from the publisher's own own websites; reproduced for comment)
- defined.ai: Data Marketplace for AI Training Datasets | Defined.ai (from the publisher's own own websites; reproduced for comment)
- withprotege.ai: Home - Protege (from the publisher's own own websites; reproduced for comment)
- withprotege.ai: For Data Providers - Protege (from the publisher's own own websites; reproduced for comment)
- snorkel.ai: Expert data development for frontier AI | Snorkel AI (from the publisher's own own websites; reproduced for comment)
- mercor.com: Mercor Data | Enterprise Workflow Data for Frontier AI (from the publisher's own own websites; reproduced for comment)
- micro1.ai: Refer a company to micro1, earn $50,000 (from the publisher's own own websites; reproduced for comment)
- innodata.com: Train and Deploy Industry‑Leading Generative and Traditional AI - Innodata (from the publisher's own own websites; reproduced for comment)
- nyne.ai: Data partnerships | Nyne.ai (from the publisher's own own websites; reproduced for comment)
- troveo.ai: Turn Your Company Data Into Revenue | Troveo (from the publisher's own own websites; reproduced for comment)
- datavendor.ai: DataVendor | Monetize your data supply (not yet read)
- simpleclosure.com: Sell Your Startup Assets | SimpleClosure AssetHub (from the publisher's own own websites; reproduced for comment)
- openai.com: OpenAI Data Partnerships | OpenAI (read, set aside)
- welocalize.com: Welocalize | Translation & Localization for Global Businesses (read, set aside)
- the universe, rule 1 (not yet read)
Use this figure
6 of 12 enterprise AI training data providers whose own websites we read say they source data by licensing it from the companies and rights-holders that own it, read 8 October 2026.
Main kind of data it supplies, by provider: licensed corpora 5 of 12, expert or RLHF data 2 of 12, data marketplace 2 of 12, annotation and labelling 1 of 12, not stated 1 of 12, varies 1 of 12 (12 providers, read 8 October 2026).
Modalities it names, by provider: stated 8 of 12, not stated 4 of 12 (12 providers, read 8 October 2026).
How it says it sources the data, by provider: licensed rights-holders 6 of 12, vetted experts 3 of 12, not stated 2 of 12, a crowd workforce 1 of 12 (12 providers, read 8 October 2026).
Workforce or network size it states, by provider: not stated 11 of 12, stated 1 of 12 (12 providers, read 8 October 2026).
Runs a programme for data owners to supply data, by provider: has a data partner programme 7 of 12, not stated 5 of 12 (12 providers, read 8 October 2026).
Citation: "Enterprise AI training data providers and what they supply", updated 2026-10-08, https://enterprisetrainingdataproviders.com/datasets/custom-data-provider-capabilities/.
The data as a CSV (70 answers, worked out 2026-10-08)