Selling direct means signing the license with the company that takes the data, while selling through a broker adds an intermediary that finds the licensee, is paid by the seller, the buyer or both, and never becomes the licensee itself. The fee claims differ by payer: DataSupply.ai states a 20 % success fee paid by the seller, and Regulr states that the buyer pays it and the business owner pays nothing (both checked 2026-11-09).
The route to the licensee is one decision in AI training data licensing, and the payer, the mandate terms and the control over data and scope matter more than the headline percentage. Not selling at all remains a valid outcome.
DataPacts Editorial Team took every term here from the named party's own pages; the page on how we source, date and check every figure sets out the rules.
What Each Route Means: Direct License or Intermediary
A direct route ends in a license with a lab, a company data program or a data vendor, and an intermediary route adds a broker, sell-side adviser, referrer or shutdown service that places the data without acquiring it. Endless and Polyshares license data from the company, so dealing with them is a direct route.
"Data broker" in the consumer and privacy-law sense means personal-data resale for marketing, which this page does not cover.
The directory of AI data licensing brokers and sell-side advisers lists each intermediary, while this page compares the routes.
Who Pays the Intermediary: Four Fee Models
Intermediaries are paid in four ways, and the payer matters more than the percentage because it decides whose interest the intermediary serves.
The table compares the fee models as of 2026-11-09; every figure is the named intermediary's own claim.
| Fee model | Who pays | Examples and their stated terms | Checked | Source |
|---|---|---|---|---|
| Seller-paid success fee, percentage published | Seller, from the license price | AIData.Broker 12 % / 10 % / 8 % / 6 % bands on broker-run sales; DataSupply.ai 20 % of an accepted offer; Pewter 10 % of the license price | 2026-11-09 (AIData.Broker pricing reviewed 2026-11-08) | aidata.broker/pricing; datasupply.ai/owners; pewterdata.com/facts |
| Seller-paid success fee, percentage not published | Seller | Corpus Bridge; Data Licensing Advisors; SimpleClosure; Avelence agrees fees separately | 2026-11-09 | Their own pages |
| Buyer-paid fee or referral payout | Licensee | aitrainer.work; Marin Data; OpenTrain AI; Regulr | 2026-11-09 (Marin terms 2026-10-17) | Their own pages |
| Spread between two licenses | Licensee of the platform | Grepped, a marketplace, splits its spread 50/50 with the referral partners who bring companies | 2026-11-09 | grepped.ai |
| No intermediary (direct) | None | Mercor no setup fees or deductions; Mode no broker fee; Polyshares no fee and no commission | 2026-11-09 | mercor.com/data; data.mode.inc; polyshares.com |
Seller-Paid Success Fees
A seller-paid success fee comes out of the license price, so the seller receives less than the buyer pays, and only three intermediaries publish the percentage (as of 2026-11-09):
- DataSupply.ai states a "20% success fee only from an offer you accept", with the buyer paying the seller directly.
- AIData.Broker states fees only on broker-run sales (pricing reviewed 2026-11-08): 12 % of the first USD 250,000, 10 % to USD 1,000,000, 8 % to USD 3,000,000, 6 % above, and its own example of USD 35,000 on a USD 300,000 sale.
- Pewter states "If a deal closes, Pewter's fee is 10% of the license price".
Data Licensing Advisors, Corpus Bridge and SimpleClosure state a success fee without a published percentage, and Avelence agrees any fee separately.
Buyer-Paid Fees and Referral Payouts
A buyer-paid intermediary charges the seller nothing and is paid by the licensee when a license closes, often from the buyer's referral program. Regulr states that the business owner pays nothing, and Marin Data states that the AI lab pays it separately from what the lab pays the company (terms updated 2026-10-17).
OpenTrain AI states that buyers crediting a referral may pay it a reward, and aitrainer.work states that some buyers pay it a fee, without giving an amount. The referral programs of AI data buyers set the pool such fees come from.
Spreads and Margins the Seller Does Not See
An intermediary that licenses the data to itself and licenses it on earns the difference between the two prices, and the seller rarely sees the second price. Grepped, a marketplace, states that it splits its spread 50/50 with the partners who bring companies to it, that licenses to buyers are perpetual, and that the seller indemnifies Grepped (terms of 2026-11-08). Endless states "No commission on your side, nothing deducted later", but it licenses on to labs, so its second price is not part of the seller's terms.
What Direct Programs Say They Deduct
Direct programs state that the price is not reduced by a broker fee, and some also carry extraction and de-identification costs. Mercor states it covers extraction, anonymization and transfer costs with no setup fees or deductions, Mode states no broker fee, and Polyshares states "There is no fee and no commission" (all checked 2026-11-09). A buyer's referral fee does not appear in the seller's price either, so net proceeds follow the method for evaluating an offer for your data.
Conflicts of Interest Between Seller and Intermediary
A conflict arises wherever the intermediary earns more from an outcome the seller would not choose, and the payer, the buyer list and the intermediary's age are the three places to look.
The table lists each check as of 2026-11-09.
| Where to look | What the published terms show | Question for the intermediary | Source |
|---|---|---|---|
| Payer | AIData.Broker: "not the agent or representative of any buyer", buyers pay referral payouts; Marin Data paid by AI labs | Who pays you, and how much from each buyer? | aidata.broker/about, /pricing; marindata.ai/terms |
| Buyer referral fees | micro1 USD 50,000 up to USD 700,000, 7.5 % above; Mode the greater of USD 50,000 or 6.00 %; Handshake USD 10K | Which buyers pay you a referral fee? | micro1.ai/company-referral; data.mode.inc; joinhandshake.com |
| Single-buyer routing | The Corner opens micro1's form with its own code | Do you approach more than one buyer? | thecorner.ai |
| Unnamed counterparties | Pewter's licensing partner and McCarthy's AI data partners are not named | Who signs the license? | pewterdata.com; mccarthys.ai |
| Track record | Regulr has not yet closed a data deal; Pewter founded October 2026; Aba Compute "5 companies live" | Have you closed a data license? | regulr.ai; pewterdata.com/facts; abacompute.com |
When the Buyer Pays the Broker
A broker paid by the buyer is paid only if that buyer signs, so its income can favor the buyer with the highest referral fee rather than the best license for the seller. AIData.Broker states that it is not the agent or representative of any buyer, that introductions are free to the seller, and that buyers pay referral payouts on introductions.
Buyer scales differ by buyer and deal size. micro1 states USD 50,000 for enterprise deals up to USD 700,000 and 7.5 % of deal value above, Mode states the greater of USD 50,000 or 6.00 % for the first to fifth qualified deal, and Handshake states "Refer a company, get $10K" paid when the deal closes (all checked 2026-11-09). No published page shows that any named intermediary steers sellers; the point is who pays.
Referral Partners That Restate One Buyer's Terms
Some landing pages that look like independent advisers route the company to a single buyer's form with a referral code. The Corner describes itself as an independent referral partner of micro1, not micro1 or its representative, and its apply button opens micro1's form with The Corner's code; micro1 may pay it a referral fee, amount not stated. McCarthy's AI arranges licensing through data partners and names neither buyers nor partners (checked 2026-11-09).
Track Record, Unnamed Licensees and Age
An intermediary's own pages often say how new it is and whether it has closed a deal, and those statements belong in the decision (checked 2026-11-09):
- Regulr states that it has not yet closed a data deal.
- Pewter states a founding in October 2026 and does not name its licensing partner.
- Aba Compute shows "5 companies live" in its footer.
Unsolicited approaches get the red flags and scam checks first.
Control: Who Signs, Who Holds the Data and Who Sets the Scope
Control in a data deal means three things: whose name is on the license, who touches the data, and who approves the scope, and published intermediary terms differ on each.
The table compares direct programs with published intermediary terms as of 2026-11-09.
| Control point | Direct program | Intermediary, terms as published | Source |
|---|---|---|---|
| Licensee | The program or lab | Seller signs with the buyer: AIData.Broker, Data Licensing Advisors; the partner signs: Pewter; buyers license under an agreement with the platform: Grepped | Their own pages |
| Custody of data | Program extracts and de-identifies (Mercor) | Never holds data: OpenTrain AI, Data Licensing Advisors; prepares a copy: Avelence | Their own pages |
| Scope approval | In the program's agreement (Mode) | Owner approves final scope: Regulr; exclusions of buyers and uses: Aba Compute | regulr.ai; abacompute.com |
| Freedom to go direct later | n/a | Non-exclusive mandate, 30 days' notice, 12-month tail limited to listed buyers: AIData.Broker | aidata.broker/pricing |
Signing Directly With the Licensee
Most published intermediary models leave the license between the seller and the licensee, so the seller keeps privity and can enforce the license without the intermediary. AIData.Broker and Data Licensing Advisors state that the holder signs directly with the buyer. Pewter differs: its unnamed licensing partner signs the license. Endless licenses the data on to labs itself, and Grepped's terms state that buyers receive rights only under a signed agreement with Grepped (terms of 2026-11-08).
Custody of the Data and the De-identified Copy
Some intermediaries never receive the data, others prepare a de-identified copy themselves or through a partner, and the seller should know which before sharing data. OpenTrain AI states that it never receives or holds data, and Data Licensing Advisors states that it never takes custody. Avelence prepares a de-identified copy under a mutual NDA, Regulr licenses only de-identified copies, and Aba Compute delivers through a de-identification partner or the buyer's controlled environment (all checked 2026-11-09).
Mandate Terms: Exclusivity, Notice and Tail Periods
A broker mandate decides whether the seller may still go direct, and the terms to read are exclusivity, the notice period and the tail period after termination. AIData.Broker publishes a non-exclusive mandate with 30 days' notice and a 12-month tail limited to listed buyers (pricing reviewed 2026-11-08). Aba Compute states that the holder can exclude named buyers, competitors, customers or uses. No intermediary in this comparison publishes an exclusive mandate term.
Exclusivity can bar a direct program: Mercor's medical-practice listing names "Data already licensed exclusively through a broker" as not a fit (checked 2026-11-09). Polyshares shows why the license text outranks the landing page: its home page says "Defined material for a defined term" while its track record says "Perpetual exclusive license to the operational record". The same lock-in logic applies as for exclusive vs non-exclusive data licenses.
Direct or Through a Broker: Benefits, Risks, Downsides and Alternatives
An intermediary adds reach and preparation work and adds a party whose pay depends on the deal closing, while going direct keeps the chain short and leaves the search to the company.
Benefits of an intermediary
- Buyer search across several programs by one party.
- Preparation on a separate copy, as Avelence states.
- Nothing charged to the seller, as Regulr states.
Risks
- A buyer-paid incentive that depends on which buyer signs.
- An exclusive brokered license that can bar a direct program.
- Young intermediaries without a closed deal.
Downsides
- A seller-paid success fee reduces proceeds; published rates run from 6 % (AIData.Broker's top band) to 20 % (as of 2026-11-09).
- A tail period can run after termination.
- One more party under NDA.
Alternatives
- Apply direct to a program such as Mercor's or Mode's.
- Compare several direct programs.
- Wait.
- Keep the data and do not sell or license it at all.
The full list is in the alternatives to selling data to AI labs, including not selling at all.
Questions to Ask Before Signing a Broker Mandate
Seven written answers separate one intermediary from another before any data or sample leaves the company.
- Who pays you and how much, by buyer?
- Is the mandate exclusive, and what notice and tail period apply?
- Who signs the license?
- Do you receive or hold the data?
- Which buyers do you approach, and may the company exclude any?
- Have you closed a data license, and for whom if disclosable?
- Is any fee or referral reward paid to anyone else?
The license itself is read clause by clause in AI data licensing agreements.
Where the Direct-or-Broker Decision Sits in the Process to Sell or License Company Data
The direct-or-broker decision belongs to step 7, choosing a buyer, after valuation and before NDAs, samples and negotiation. The 14 steps to sell or license company data to AI companies start with whether to sell or license data at all.
- Previous: how much company data is worth
- This step: choosing a buyer for your data
- Next: running several AI data buyers in parallel, NDAs before sharing data with an AI buyer, PII removal and de-identification, negotiating an AI data licensing deal and when and how AI data buyers pay
- Related: AI training data marketplaces and dataset vendors and data labs
Intermediaries for Closed and Winding-Down Companies
Closed companies meet the same choice, and shutdown services add a third route that bundles the data sale with the wind-down. SimpleClosure states "no upfront cost" and a percentage only on success, Pewter accepts winding-down or closed companies, and Marin Data states that winding-down companies may qualify (all checked 2026-11-09).
The page on selling data from a defunct company follows the same choice. A trustee or administrator decides for data assets in insolvency, and shutdown services for closing companies are listed separately.
Intermediaries and Programs Cited on This Page
The terms above come from each party's own pages, checked November 9, 2026 unless dated otherwise. Intermediaries and licensees: Aba Compute, AIData.Broker, Avelence, Corpus Bridge, Data Licensing Advisors, DataSupply.ai, Endless, Grepped, Marin Data, McCarthy's AI, Mode, OpenTrain AI, Pewter, Polyshares, Regulr, SimpleClosure, The Corner and aitrainer.work.
Buyer programs: company data partnership programs, the Mercor data buyer profile, the micro1 data buyer profile, the Handshake AI data buyer profile and AI data partnership payouts and program terms compared.
Frequently Asked Questions
Do I pay a broker if the buyer pays its fee?
Not directly; a buyer-paid broker such as Regulr or Marin Data states the company pays nothing, but the fee is part of what the buyer spends on the deal.
Does a broker get a better price than going direct?
No public figure shows it; no seller-confirmed comparison of brokered and direct prices exists. The fee of a seller-paid broker, such as DataSupply.ai's 20 % (checked 2026-11-09), reduces proceeds by that share.
What is a broker mandate agreement for data licensing?
A mandate is the contract that authorizes an intermediary to approach buyers for the company and sets its fee, exclusivity, notice and tail period. AIData.Broker, for example, publishes a non-exclusive one.