AI Training Data Marketplaces: Fees, Eligibility and Payouts Compared

An AI training data marketplace is a platform where data holders list or license datasets, records or media and AI developers obtain them, with the platform charging a fee or keeping a share of each sale. The directory below covers 19 such marketplaces, from self-serve listing sites such as AWS Data Exchange and Opendatabay to managed platforms such as Troveo and contributor apps such as Kled. Most self-serve marketplaces publish a percentage they keep, from 1.5 % to 30 %, while managed platforms publish a revenue split or nothing, and two publisher libraries publish no rate at all.

This page is part of the independent reference on AI training data licensing: the order is alphabetical, every figure is the marketplace's own claim checked 2026-11-09, and a marketplace is listed whether or not it runs a referral program. DataPacts is not a broker and takes no cut of any sale.

Each number carries its source and check date, and how we source, date and check every figure is set out in the methodology.

What an AI Training Data Marketplace Is

In the AI training sense, a marketplace stands between a data holder and an AI developer, and in most cases it never owns the data: the seller lists it, or the platform licenses it on the seller's behalf. DataVendor states that a sale "licenses your assets for a commercial use rather than transferring them", and Troveo states that the seller keeps ownership while Troveo licenses on its behalf, both checked 2026-11-09. Kled is the exception in this directory, because its terms of service take a perpetual, irrevocable license that is exclusive to Kled for training third-party AI models.

The generic sense of "data marketplace" is wider than this directory, which covers only platforms that list, license or distribute data for AI training. AWS Data Exchange, Databricks Marketplace and Snowflake Marketplace are included because they accept third-party datasets from providers. A marketplace also differs from a company data program, which is a standing offer by one buyer to license a company's records, and from a broker, which negotiates for a seller.

In the chain to AI labs, a marketplace is one intermediary among several, and the page on who buys AI training data shows the others. Defined.ai, Pangea.ai, Protege, HUMXN and ProRata are also typed as marketplaces in the reference, but they sit under the program and platform categories.

Every AI Training Data Marketplace on Record

The directory lists the 19 marketplaces profiled in this category, alphabetically, with the operator each one names and the model it runs. The table is as of 2026-11-09, and every row carries its check date and source.

• "Model" is the class defined in the next section.
• "What it acquires or handles" never means that the platform buys the data.
• "not published" in the Operator column means no primary page names the operating entity.

Four Marketplace Models: Self-Serve Listing, Managed Licensing, Contributor Apps and Publisher Libraries

Marketplaces differ less in the data they take than in who does the work and who signs with the AI buyer. The reference sorts the 19 into four models.

• Self-serve listing (8): the seller lists, sets terms and signs with the buyer. AWS Data Exchange, Databricks Marketplace, Datarade, DataVendor, FileYield, Mozilla Data Collective, Opendatabay and Snowflake Marketplace.
• Managed licensing (6): the platform prepares the data and licenses it on the holder's behalf. Created by Humans, Grepped, Human Native, Magma, Newsflare and Troveo.
• Contributor apps (3): individuals upload and the platform licenses to AI buyers. Kled, Luel and Wirestock.
• Publisher libraries (2): a stock library licenses its own collection and pays contributors. Getty Images and Shutterstock.

Managed licensing resembles the company data partnership programs in that the platform signs with the buyer, but a program is run by one data company for its own purchases or placements and a managed marketplace offers many holders' data to many buyers. Grepped, for example, states that it prepares, contracts and delivers, and Troveo states it has 40+ active buyers, checked 2026-11-09.

The reference types Wirestock as a human-data platform and Getty Images and Shutterstock as publishers, and they sit here because each licenses content to AI developers and pays the people who made it.

Marketplace Fees Compared: Listing Fees, Commissions and Revenue Splits

Most self-serve marketplaces publish a percentage they keep, managed platforms publish a revenue split or nothing, and the two publisher libraries publish no rate at all. The table is as of 2026-11-09, and each cell is the marketplace's own claim.

Chart note: FileYield's seller fee is shown at 8 %, and its how-it-works page says listing is free.

Listing Fees and Commissions on Self-Serve Marketplaces

AWS Data Exchange states a listing fee of 3 % on public offers and lower tiers on large private offers, while Datarade and Opendatabay take commissions that fall as the seller pays more or sells more. AWS states 3 % on private offers under USD 1 million, 2 % from USD 1 million to under USD 10 million and 1.5 % from USD 10 million, with renewals at 1.5 %, effective 2024-01-05. Datarade states a commission-only plan at 30 %, Bronze at USD 6,000 a year plus 20 %, Silver at USD 12,000 a year plus 15 % and a custom Gold plan, checked 2026-11-09. Opendatabay states a free listing and a commission of 30 % on sales of USD 1 to 100 falling to 5 % from USD 250,001, so the seller keeps 70 % to 95 %, under a schedule effective 2026-10-06.

DataVendor states that it keeps 20 % of vendor earnings, assessed only when earnings are withdrawn, with a USD 1,000 minimum withdrawal and a USD 15 fee when a chargeback is lost, checked 2026-11-09. Snowflake Marketplace states no fee of its own, and Mozilla Data Collective states that uploaders "keep 100% of revenue", both as claims checked 2026-11-09. Databricks Marketplace states no fee. A seller weighing a commission-based directory against a cloud listing can read the side-by-side comparison of Datarade vs AWS Data Exchange.

Revenue Splits on Managed and Contributor Marketplaces

Managed platforms and contributor apps pay the rights holder a share of what AI buyers pay, and only Newsflare and Wirestock publish the share. Newsflare states a "50:50 revenue share on AI licensing", paid on a monthly cycle with the creator's portion sent within 30 days of AI clients' usage reports, and Wirestock states "You earn 50% commission each time someone licenses one of your assets", plus fixed payments per approved asset, both checked 2026-11-09. Magma states a revenue share whose percentage is set in onboarding, Troveo states a royalty per use and "no public price sheet" (2026-07-25), Kled pays rewards in fiat or tokens such as SOL at its discretion, and Created by Humans and Luel publish no share.

Grepped's 50/50 is an introducer split, not a seller fee: it states that partners "split the spread" with it, and its fee to a data holder is not stated. Splits of this kind are covered with the referral programs of AI data buyers. Getty Images and Shutterstock publish no rate. Getty's FY2025 10-K, filed 2026-03-16, states only that contributors are compensated when their works are used as training data.

Where a Marketplace's Own Pages Disagree

Four marketplaces publish figures that contradict each other on their own pages, and DataPacts shows both values rather than picking one. All four were checked on 2026-11-09.

• FileYield: "Listing on FileYield is free" on its how-it-works page, against "8% for sellers and 6% for buyers" in the sell-data FAQ.
• Datarade: 2,700+ providers on its home page, against 600+ providers on its provider apply page.
• Troveo: USD 50M+ paid to creators on its content-owners page, against USD 20M+ paid to licensors on its business-data page (checked 2026-11-09).
• Opendatabay: a payout "[30]" days after month end with a minimum "[$50]", where the brackets appear on the fee page itself (schedule effective 2026-10-06).

Each value is a claim by the marketplace, and none is a total. No seller has publicly confirmed an amount received through any marketplace in this directory as of 2026-11-09, which is the gap that advertised vs confirmed payouts tracks.

Eligibility: Who May List and Which Data Marketplaces Refuse

Most self-serve marketplaces admit registered businesses only, and several refuse personal data outright. The table is as of 2026-11-09, and each cell is the marketplace's own claim.

Datarade states that it onboards legally registered businesses only and that individuals and unregistered businesses are not onboarded, while its application takes about 3 minutes and review usually 1 to 2 business days. Snowflake states that USD payouts need a US billing address, and Databricks requires a Premium plan or above. Datarade prohibits unanonymized personal data and illegally scraped data, and AWS excludes data identifying a person outside its Extended Provider Program. Every listing assumes the seller holds the rights, and the rights audit is the step that establishes whether it does. Troveo states that content owners must own or hold rights to license the content, and DataVendor excludes public repositories.

In the EU, Regulation (EU) 2022/868 (the Data Governance Act) has been in force since 2022-06-23 and applies from 2023-09-24 (Art. 38). Data intermediation services must notify a competent authority (Art. 11(1)) and meet the Art. 12 conditions, and non-EU providers designate a legal representative (Art. 11(3)). Whether a given marketplace is such a service depends on what it does, and the Commission's proposal COM(2025) 837 would repeal the Act.

Payout Triggers and Timing on Marketplaces

Marketplace money reaches the seller only after the AI buyer has paid, and each marketplace states its own trigger: a monthly disbursement, a withdrawal, a usage report or each sale. The triggers below are claims checked 2026-11-09.

• AWS Data Exchange: a monthly disbursement less fees.
• Snowflake Marketplace: Stripe pays providers after receiving payment from consumers.
• Opendatabay: payout to a verified business account after month end, with the day count and minimum shown in brackets on its fee page.
• DataVendor: earnings are paid out on withdrawal, with a USD 1,000 minimum (checked 2026-11-09).
• Newsflare: a one-month cycle, with the creator's portion sent within 30 days of AI clients' usage reports.
• Wirestock: per license for commission projects, and "instantly" for fixed payments per approved asset.

Troveo states that it pays "on every sale", FileYield states that the parties handle payment and delivery themselves with no escrow, and Datarade, Databricks and Mozilla Data Collective state no trigger. The reference on when and how AI data buyers pay covers milestones, revenue share and clawbacks, and none of these pages states a deadline that a seller can enforce.

The License a Marketplace Passes to the AI Buyer

What a seller gives up depends less on the fee than on the license: some marketplaces leave the seller as licensor, others take an exclusive, perpetual grant before any AI buyer appears. On Datarade, Snowflake Marketplace and Databricks Marketplace the provider sets its own terms with the buyer, AWS offers an optional Data Subscription Agreement template or custom terms, and DataVendor states that its buyer's entitlement survives the seller unpublishing the listing.

Broader grants appear on the contributor side. Kled's terms of service, revised 2026-07-21, grant Kled a transferable, sublicensable, worldwide, perpetual and irrevocable license that is exclusive to Kled for training third-party AI models, and state that Kled may but is not obliged to delete on account deletion. Newsflare requires footage to be exclusive to it for AI licensing and to stay 90 days after upload, with deletion on request taking 30 days. Grepped's terms (2026-11-08) state perpetual licenses to buyers, exclusive or not, with the seller indemnifying. Troveo states that 95 % of its licensors signed exclusively in a resource dated 2026-07-25, although its program pages do not state exclusivity. Magma requires a separate written definitive agreement.

Deletion rarely undoes a license already granted, so the clause to read first is the one on what survives removal. The AI data licensing agreements page explains exclusivity, term and deletion clauses one by one.

Ownership Changes, Insolvency and Revenue Shifts Among Marketplaces

Three corporate events since January 2026 changed who runs or funds marketplaces in this directory: an acquisition, a terminated merger and a chapter 11 filing. Cloudflare announced on 2026-01-15 that Human Native joins Cloudflare, with the price not disclosed, and Human Native has no current program terms on record. The program terms Cloudflare publishes for publishers sit with Cloudflare and Human Native.

Getty Images filed a prepackaged chapter 11 case on 2026-10-21 in the US Bankruptcy Court for the District of Delaware, and no outcome is stated here. The merger agreement of Getty Images and Shutterstock, dated 2025-01-06, was terminated on 2026-07-07, so the two companies are not merged.

Shutterstock reports Data, Distribution and Services revenue of USD 15.823 million for Q3 2026 against USD 65.668 million for Q3 2025 (10-Q, period ended 2026-09-30), and USD 203.264 million for FY2025 (10-K), with no AI-only line. Funding claims are the marketplaces' own: Kled states a seed round of USD 5.5 million and USD 10 million in total, Luel states a seed round of USD 31.2 million led by General Catalyst and Lightspeed (2026-05-15), and Wirestock states a USD 23 million round, all checked 2026-11-09.

Benefits, Risks, Downsides and Alternatives of Listing on a Marketplace, Including Not Selling

A marketplace gives a data holder reach to AI buyers without negotiating each deal alone, but the fee, the license and the platform's own future come with it. The lists below are neutral and do not rank any marketplace.

Benefits:

• Reach: Snowflake states 820+ providers and AWS states 3,500+ products and 300+ providers, both checked 2026-11-09, so a listing sits in a catalog buyers already use.
• Ownership kept: DataVendor states that a sale is a license, and Troveo states that the seller keeps ownership.
• Process: Datarade states review usually takes 1 to 2 business days, and Troveo states it licenses on the holder's behalf.

Risks:

• License breadth: Kled takes an exclusive, perpetual, irrevocable license, and Grepped's terms make the seller indemnify the buyer.
• Contradictory terms: FileYield's own pages disagree on whether listing is free, and its terms page is marked a draft not reviewed by counsel.
• Platform events: a platform can be acquired or enter chapter 11, as Human Native and Getty did, and the list of data buyer incidents records disputes and breaches.
• Unconfirmed payouts: no seller has publicly confirmed an amount received.

Downsides:

• Fees of 1.5 % to 30 %, charged on every sale or at withdrawal.
• Annual plan fees, such as Datarade's USD 6,000 or USD 12,000 a year (checked 2026-11-09), before any sale.
• Eligibility limits: businesses only on Datarade, DataVendor and AWS, and a US billing address for USD payouts on Snowflake.

Alternatives:

• A company data program that licenses the records on one published process.
• A broker or sell-side adviser that negotiates for the holder.
• A direct license to an AI lab under a negotiated contract.
• Not selling or licensing at all, which is a valid outcome and costs nothing.

Whether to list at all turns on the data, the rights and the alternatives, which the guide on whether to sell or license data sets out in full.

Where Marketplaces Sit Among Programs, Brokers, Vendors and the Seller Path

A marketplace is one of several routes by which data reaches AI developers; the pages below cover the others and the seller's steps around a listing. The seller path runs from readiness to payment, and the disclosure page sets out how DataPacts is funded.

• Choosing between routes: choosing a buyer for data, and sell or license company data to AI companies.
• Terms: program terms compared and PII removal and de-identification.
• Other intermediaries: AI data licensing brokers, dataset vendors and data labs and content licensing platforms for publishers.
• Seller steps: how to sell a dataset and alternatives to selling data.
• Deals and comparisons: AI data licensing deals, Troveo vs Protege and data intermediation services under EU law.

This page is information, not legal advice.

Questions About AI Training Data Marketplaces

What is a data marketplace?

A data marketplace is a platform where data holders offer data and buyers obtain it under a license, with the platform taking a fee. In this reference the term covers 19 platforms that list, license or distribute data for AI training.

Which AI training data marketplace is best?

No marketplace is best for every seller; fit depends on the data, the seller's country and size, and the license the seller accepts. DataPacts does not rank them.

Is a data marketplace the same as a data broker?

No: these marketplaces list or license data that holders choose to offer for AI training, while consumer data brokers trade personal data for marketing, which this site does not cover.

Can individuals sell their own data on these marketplaces?

Some can: Kled, Luel, Wirestock and Newsflare take uploads from individuals, while Datarade, DataVendor and AWS Data Exchange admit organizations only. Paid tasks for individuals are compared under AI training jobs and platforms.

Are enterprise data catalog marketplaces the same thing?

No: internal data catalog marketplaces share data inside one company and are not a route for licensing data to AI developers.