- A data provider supplies data under license again and again, so buyers check the supplier as well as the data.
- Due diligence covers where the data came from, the rights chain, personal data and notice, security and quality.
- Providers usually price a delivery stream: a subscription, a fee per delivery or usage fees, with refresh as the main lever.
- Exclusivity with one AI lab turns a provider back into a one-time seller; keep any grant narrow, short and paid for.
- Channels are direct, cloud marketplaces, company data programs, AI data marketplaces and independent partners; judge each by who signs and who sets the price.
What do AI buyers ask for before you become a data provider?
A data provider sells access to data again and again, so buyers check the supplier as well as the data. The table lists what each kind of buyer or rule-setter asks of a provider, taken from its own published page. Every row was read on 20 November 2026.
| Buyer or rule-setter | What it asks a provider for | Source | Date checked |
|---|---|---|---|
| Cloud marketplace: AWS Data Exchange | Data that meets AWS Marketplace legal eligibility; nothing that identifies a person unless publicly available; accurate descriptions that say if data was aggregated or anonymized; a subscription agreement that sets permitted uses | AWS Data Exchange publishing guidelines and provider guide | 20 Nov 2026 |
| Cloud marketplace: Snowflake | A usage-based or subscription-based plan for paid listings, and a trial for any paid listing offered publicly | Snowflake paid listings documentation | 20 Nov 2026 |
| Cloud marketplace: Databricks | Agreement to its provider policies and an account on the Premium plan or above | Databricks provider documentation | 20 Nov 2026 |
| Company data program: Scale AI | Real operational workflows, not demos or synthetic data; structured, de-identified records; a clear owner who can approve packages; 40 or more employees preferred | Scale AI data partnership page | 20 Nov 2026 |
| Large buyer security review: Microsoft | Suppliers working with personal data, confidential data or AI systems meet its Supplier Data Protection Requirements, and new suppliers cannot start until the compliance cycle is complete | Microsoft Supplier Security and Privacy Assurance page | 20 Nov 2026 |
| Standards body: Data & Trust Alliance | 22 provenance fields covering source, provenance and use, including where consent documentation is kept and the license to use | Data Provenance Standards v1.0.0 | 20 Nov 2026 |
| Regulator: European Commission (AI Act) | General-purpose AI providers publish a summary of training data that lists main data collections and explains other sources | Commission template announcement | 20 Nov 2026 |
We work only with data you created in running your business, or hold under a written licence that lets you license it for AI training. We do not take scraped web data, purchased contact or consumer lists, people-search or location data, or personal data collected without a notice that covers this use. Every row in the table is easier to answer for data you made yourself, because you can show each step from creation to delivery.
What is a data provider?
A data provider is a company that supplies data to others under a license, usually as a repeating delivery rather than a one-time file. AWS Data Exchange uses the word for anyone who registers to publish data products there, and its providers keep data sets current by publishing new revisions (AWS). In AI, a data provider supplies the records that models are trained or tested on. A data vendor is usually the same thing under another name.
Three neighbors are easy to confuse with it. A data aggregator collects data from many sources and licenses the combination; Defined.ai, for example, describes its data partners as any business that generates or aggregates data, and keeps collection and annotation firms in a separate track. A data broker, in California law, is a business that knowingly collects and sells to third parties the personal information of a consumer with whom it does not have a direct relationship (Cal. Civ. Code 1798.99.80). A collection or annotation vendor is paid to make new data to a buyer's order. This page is for the first kind only: a first-party provider licensing data it created in running its business.
How do I become a data vendor to AI companies?
Treat it as a product launch, not a one-off sale. The steps below follow from what buyers' pages and marketplace rules require, in the order that avoids rework.
- Scope the product. Name the systems, record types, date range and volumes, and what will be delivered each month or quarter.
- Clear the rights. List who created each part and under which contracts, plus any third-party content inside. Drop what you cannot clear.
- Handle personal data. Check the notice it was collected under, decide what to remove, and record how you removed it.
- Document it. Write a dataset card, a data dictionary and a provenance sheet that follows the Data & Trust Alliance fields.
- Prepare for security review. Have answers ready on access control, encryption, retention, and where the data is stored and processed.
- Set your terms. Decide permitted uses, term, refresh cadence, exclusivity, support response times and what happens when the license ends.
- Choose channels. Decide which buyers you approach directly, which marketplaces you publish on, and whether a partner works for you.
Large buyers run formal supplier reviews, and the larger the buyer, the more forms it sends. A small team can still pass if one person owns the answers. Scale AI, for one, asks for a clear owner who can scope the partnership, approve packages and keep reviews moving. AWS Data Exchange asks providers of API products to answer subscriber support questions within one business day.
What due diligence do AI labs run on data vendors?
Expect questions in five groups. Few AI labs publish their checklists, so the groups below come from the published standards, rules and buyer pages that such reviews follow. Keep the evidence for each in one folder before the first call.
- Where the data came from. How it was created, when, by whom and by what method. The Data Provenance Standards were written for this; 19 member organizations of the Data & Trust Alliance developed them, and IBM tested them in its clearance process for datasets used to train foundation models.
- Rights chain. Who owns each part, and the license terms on any third-party content. In the EU a general-purpose AI provider publishes a summary of its training data that lists main data collections and other sources, and the Commission says the summary helps copyright holders exercise their rights (European Commission). Your rights file feeds the buyer's own disclosure.
- Personal data and notice. What personal data the set holds, the notice people saw when it was collected, and how it was removed. The UK ICO has said it found a serious lack of transparency about training data in the AI industry, and that where poor transparency leaves people unable to exercise their rights, generative AI developers are likely to struggle to pass its balancing test (ICO). A careful buyer therefore asks its suppliers first.
- Security. How the data is stored, moved and deleted. Microsoft, for example, applies its Supplier Data Protection Requirements to suppliers working with personal data, confidential data and AI systems, and says new suppliers in scope cannot start work until the compliance cycle is complete (Microsoft).
- Quality and fit. A sample, a data dictionary, counts by year and field, known error rates, and how any labels were made.
Most agreements also carry a warranty that you hold the rights you license, and an indemnity if that proves wrong. Read that clause against your rights file before you sign. The data licensing due diligence checklist and the data licensing agreement guide go clause by clause.
How do data vendors price datasets for AI training?
A provider prices a delivery stream, not one file, so the unit is usually time or use: a monthly or yearly subscription, a fee per delivery, or usage fees on a marketplace. Snowflake's paid listings, for example, use either a usage-based plan or a subscription-based plan (Snowflake). On AWS Data Exchange a provider can publish a public offer and also make private custom offers to selected customers.
Company data programs publish ranges instead of prices. Scale AI gives an illustrative value of $10K to $1M+ per data partnership, scaling with cadence. Handshake AI says qualified partnerships typically run from $100K to $4M, depending on the data. These are ranges a buyer advertises before it has seen your records, not offers, and no AI lab we checked publishes a unit price for supplier data.
Refresh is where providers earn more than one-off sellers. A dataset that grows each month can carry a recurring fee, because the buyer pays for freshness as well as volume. Agree the cadence, the format of each delivery and what counts as a late or short delivery. The pricing models guide compares flat, recurring and revenue share terms.
Should a data vendor go exclusive with one AI lab?
Usually not for the whole product. Exclusivity turns a provider back into a one-time seller: one buyer, one fee, and no second sale until the term ends. It can make sense when the lab pays clearly for it, through a higher fee, a minimum guarantee, or a funded collection project that would not happen otherwise.
If you grant it, keep it narrow and short. Limit it to one field of use, one data slice or one period, give it an end date, and keep the right to use your own data inside your business. Check every existing license first, because a marketplace listing or an earlier non-exclusive deal may already rule out an exclusive grant. The guide to exclusive and non-exclusive licenses covers the drafting.
What is the best channel for a data vendor to reach AI labs?
There is no single best channel; each suits a different provider. The table compares them on what you control and who does the work, as of 20 November 2026.
| Channel | How it works | Fits a provider that | What to know |
|---|---|---|---|
| Direct to an AI developer | You approach the lab and negotiate one license | Has a rare dataset and a team to run diligence | Few labs publish a supplier door. OpenAI's Data Partnerships page said on 20 November 2026 that it is not accepting submissions |
| Cloud data marketplace | You publish a product to subscribers on one cloud | Has structured data many buyers can use as it is | Each marketplace sets its own eligibility, personal data and pricing plan rules |
| Company data program | A data company licenses your records and supplies AI developers | Has operational records and wants packaging handled | Programs publish ranges and size preferences, not prices |
| AI data marketplace | The marketplace sells licenses to your data to its AI customers and pays you licensing fees from sales | Has a ready audio, video, image or text dataset | Check who sets the price and whether the marketplace may resell |
| Independent data partner | Works on your side on rights, packaging and terms, and puts the data in front of more than one buyer | Wants help but wants to sign its own license | Check how the partner is paid and whether it takes any rights in your data |
Data provider services and data vendor lists exist for every route, but a place on a list is not a buyer. Judge each channel by who signs the license, who sets the price and who holds your data in between.
Can a data vendor work with an intermediary to reach AI buyers?
Yes, but check four things first. Who pays the intermediary, and whether that payment depends on which buyer wins. Whether it takes title to your data, a license to resell it, or no rights at all. Whether it asks for exclusivity over your data or your buyer contacts. And whether you see and sign the final license yourself. The comparison of services that help companies sell data sets out how each type is paid.
We help you license your data. Origin Data Partners is an independent data partner that helps business owners value and license their company records: the company pays Origin nothing, and a buyer may pay Origin when a deal completes. Run the same four checks on us: a buyer, not you, may pay Origin, and only when a deal completes; Origin never takes title to your data or a right to resell it, and is not a marketplace or an exchange; and the license is between your company and the buyer, signed by you. If you created the data in running your business, the two-minute estimate asks the questions on this page, and no file has to leave your systems to answer them.
Not the owner? If a client or employer holds a dataset like this, advisers can introduce the company.
Annotation platforms can be one more kind of intermediary: SuperAnnotate prepares a company's records and relicenses them to AI teams, with the company approving each buyer (our SuperAnnotate Data Partnerships review).
How we checked this
We read every source on this page on 20 November 2026. The primary rules were a statute (California Civil Code 1798.99.80), a regulator's guidance (the UK ICO), the EU's official AI Act template announcement and a published industry standard (the Data & Trust Alliance). The market facts come from the official provider documentation of AWS Data Exchange, Snowflake and Databricks, a large buyer's supplier rules (Microsoft), and the own pages of Scale AI, Handshake AI, Defined.ai and OpenAI. Each fact is quoted word for word in our evidence file.
What we could not verify: the checklists AI labs use in private reviews, which are not public; the prices in any provider agreement; and whether a named program is open to a given provider today. Two FTC staff posts on AI training and privacy commitments that we looked for no longer load, so they are not cited.
What to prepare in your first week as a provider
Make a one-page provider sheet: data types, systems, date range, volumes, refresh cadence, the rights you hold, the personal data involved and how it is handled, and the security controls in place. Attach the dataset card and data dictionary. A buyer can screen you from that page, and you learn early which questions you cannot answer yet.
Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.
Common questions
What is the difference between a data provider and a data aggregator?
A first-party data provider licenses data it created itself, such as records from running its own business. A data aggregator gathers data from many sources and licenses the combination, so its rights depend on every upstream license. Buyers review the two differently: a provider shows one chain from creation to delivery, an aggregator has to show many, and each link can fail.
Is a data provider the same as a data broker?
Not by the usual legal meaning. California defines a data broker as a business that knowingly collects and sells to third parties the personal information of consumers it has no direct relationship with. California is one of several states with a broker registry; our data broker guide covers the others. If your data holds consumer personal data you did not collect directly, take advice before you license it.
Will an AI buyer run a security review on a data provider?
Expect one, especially if the data held personal or confidential information. Microsoft, for example, applies its Supplier Data Protection Requirements to suppliers working with personal data, confidential data and AI systems, and new suppliers in scope cannot start work until that cycle is complete. Have written answers on access control, encryption, retention, deletion and where data is stored.
How often should a data provider deliver updates to an AI buyer?
As often as the data meaningfully changes and the buyer will pay for. Monthly and quarterly deliveries are common choices for operational records. On AWS Data Exchange providers update a data set by publishing a new revision. Whatever the cadence, write down the format of each delivery, what counts as late or short, and what happens if a delivery fails.
Can a small company become a data provider to AI companies?
Yes, but programs differ. Scale AI states 40 or more employees as its preferred starting point for a data partnership. The cloud marketplace documents we read set account and policy requirements rather than a headcount floor. What matters more than size is a dataset others cannot easily rebuild, clean rights, and one person who can answer a buyer's questions quickly.
Can a data provider license data it bought from another company?
Only if the written license it holds allows sublicensing for AI training, and buyers will ask to see that chain. Many purchased data licenses forbid it or limit use to internal analytics. Origin works only with data you created or hold under a written license that lets you license it for AI training, so purchased lists and resold data are out of scope.
What should a data provider agreement with an AI company include?
The data in scope and its versions, permitted uses such as training or evaluation, term and territory, exclusivity if any, delivery cadence and format, fees and when they are paid, security and privacy duties, warranties on rights with an indemnity, and what happens to copies at the end. Read each clause against your rights file before signing, with your own counsel.
Sources and further reading
- AWS Data Exchange User Guide: Publishing guidelines
- AWS Data Exchange User Guide: Providing data products on AWS Marketplace
- Snowflake documentation: Paid listings pricing models
- Databricks documentation: Become a Databricks Marketplace provider
- Data & Trust Alliance: Data Provenance Standards v1.0.0
- European Commission: Template for general-purpose AI model providers to summarise training data
- ICO blog: Generative AI developers, it is time to tell people how you are using their information (12 December 2024)
- California Civil Code section 1798.99.80 (data broker definition)
- Microsoft Procurement: Supplier Security and Privacy Assurance Program
- Handshake blog: Introducing Handshake AI Employer Data Partnerships
