Mercor asks for firm-owned time series and "the right to license it for AI training", and lists redistributed exchange feeds as not a fit.
NYSE's policy document of 2022-03-21 contains no wording on AI or model training, and the other venues and vendors read for this page publish no AI training license either, as of 2026-11-17. This page covers market data for AI training data licensing and is information, not a conclusion about any agreement.
Figures on this page are claims by the stating party, and policy and rule texts were read on 2026-11-17. See how we source, date and check every figure.
Market Data a Holder Owns and Market Data It Only Receives
A holder can license market data it generated itself, such as its own time series, and cannot relicense feeds it receives from an exchange or vendor, which Mercor lists as not a fit. The table sorts four kinds of market data by whose right decides licensing, as of 2026-11-17.
| Kind of market data | Whose right decides licensing | On the buyer's or exchange's own wording | Source |
|---|---|---|---|
| Time series, order-book recordings and risk data the firm generated | the firm's own right to license it for AI training | Mercor: requirement "that the firm owns" and "The right to license it for AI training" | api.data.mercor.com |
| Quotes and trades received from an exchange feed | the exchange or vendor agreement | Mercor: "Redistributed exchange feeds the firm doesn't own" not a fit; NYSE: use categories and derived data wording, no AI wording | api.data.mercor.com; nyse.com |
| Vendor reference data and analytics | the vendor agreement | no vendor document read; nothing stated | n/a |
| Data already public | nothing to license | Mercor: "Data that is already public" not a fit | api.data.mercor.com |
The right to license a compiled dataset can also turn on copyright or the database right in a dataset, which has its own page; this page makes no statement on whether prices or quotes are protected.
Reference data and vendor analytics are a separate case, since no vendor document read for this page says who may license them for training.
What an Exchange's Non-Display Policy Says: NYSE's Own Text
NYSE's document "NYSE Proprietary Market Data", dated 2022-03-21 and published by Intercontinental Exchange, defines Non-Display Use of real-time NYSE Market Information as "accessing, processing or consuming" it "for a purpose other than in support of a further internal or external redistribution" (s.2.1).
The table sets out what the sections say and what they leave open for AI use, as read on 2026-11-17.
| Section | What the text says | Why it matters for AI use | Source |
|---|---|---|---|
| 2.1 Non-Display Use definition | "accessing, processing or consuming" NYSE Market Information for a purpose other than further redistribution | the text does not say whether model training is processing or consuming | nyse.com |
| 2.2 Categories | Category 1 own behalf and Category 2 on behalf of clients, each "including the creation of derived data"; Category 3 internal matching of orders | derived data is named for internal use only; weights are not mentioned | nyse.com |
| 2.3 Examples | high frequency or algorithmic trading, order generation, investment analysis, risk management, compliance, portfolio management | trading and operations uses; no AI example | nyse.com |
| 2.4, 2.5 Reporting and declaration | no reporting requirement; an Exhibit A declaration before a feed is authorized, then annually and on any change of use | each recipient declares its use, and a change of use needs a new declaration | nyse.com |
| 3.2 Academic pricing | use limited to faculty and students; "will exclude any use" for "commercial purposes" | a grant is tied to a stated use | nyse.com |
The text does not say whether trained weights count as derived data, whether a license survives termination, or whether any AI use is allowed or barred; the text read contains no wording on AI or model training.
The academic clause is the clearest contrast: a license is granted for a stated use, not for a dataset as such.
The document says it does not alter the rights and obligations in agreements, so the vendor or subscriber agreement controls over this description. NYSE's pricing, policies and contracts page, read 2026-11-17, lists a "NYSE Proprietary Data Non-Display Declaration Form" among its documents.
Training License Versus Access License: Questions the Contract Must Answer
An access license gives the right to receive and use data for a stated purpose, while a training license, as this page uses the term, would name model training as a permitted use; the two are different grants, and silence in an access license is not a grant of the second.
These are this page's own terms and not NYSE, LSEG or Nasdaq terms, and the NYSE text read names no training license. The scope of any grant is the subject of the page on field of use.
Six questions separate the two grants in a market data agreement:
- Does the agreement name model training or machine learning, within its field of use and scope limits?
- Is the use display, non-display (and which category) or redistribution?
- Are model weights or outputs derived data, and who may keep them, a matter of ownership of derivatives and models?
- Must data or derived products be deleted at termination?
- May the model be shared, served to clients or sublicensed, given that Category 2 covers use on behalf of clients?
- Is a declaration or audit duty triggered by the use?
The list is a set of questions for the contract and the vendor, not a conclusion about any agreement. Commentary on these questions by vendors and intermediaries is not used as a source here.
What Other Venues and Vendors Have on Record
On 2026-11-17 no exchange or vendor page read for this page publishes a license for AI model training: NYSE's policy has no AI wording, LSEG's data page uses "AI-ready" as marketing without license terms, Nasdaq's pages read state none, and Bloomberg's page could not be read.
The table lists the four in alphabetical order; "nothing stated" means nothing on AI training license terms was found in the text read, not that AI use is allowed or barred.
| Venue or vendor | Page or document read, date read | What it states on AI | Source |
|---|---|---|---|
| Bloomberg | professional data pages; HTTP 403 on 2026-11-17 | not read; nothing recorded | bloomberg.com |
| LSEG | "Data & Feeds" page, title ending "Market Data Feeds and AI-Ready Data", 2026-11-17 | marketing: "AI-ready workflows at enterprise scale"; "licensing clarity built in"; no license document or training terms on the page | lseg.com |
| Nasdaq | nasdaq.com data solutions and Nasdaq Data Link pages, 2026-11-17; the Nasdaq Trader market data policies page returned "Page Not Available" | nothing stated on AI training licenses | nasdaq.com |
| NYSE | "NYSE Proprietary Market Data", 2022-03-21; contracts page 2026-11-17 | no AI or model-training wording; a Non-Display Declaration Form is listed | nyse.com |
Vendors that offer an AI training license explicitly are not named here, because no vendor document read for this page states one. A vendor's marketing line is not a license, and a written answer from the vendor is the only record that settles the question.
Rule 603(a)(2) of Regulation NMS requires national securities exchanges and associations to distribute quotation and transaction information on terms that are "not unreasonably discriminatory" (17 CFR 242.603(a)(2), eCFR as of 2026-11-13, read 2026-11-17). It limits the terms for NMS stock data, and the text read says nothing on AI use.
Licensing Market Data You Own: Time Series, Order Books and Reference Data
A firm can offer AI buyers market data it generated, such as its own order-book recordings, risk data and trading records, if it owns the series and can show the right to license it for training.
Mercor's listing "Hedge Fund or Quant Firm", published 2026-10-17, summarizes the ask as "License proprietary market, risk and sports data, or the code that trades on it", with the scoping questions "Which time series do you own, and how far back?" and "Can you license it for model training?".
Mercor states a deal range of USD 25,000 to 1,150,000, "Depends on history and exclusivity", re-read 2026-11-17; it is a claim for a whole deal and not a price per record. The page on investment firms and hedge funds licensing market data covers the listing in full, with material nonpublic information and deal records.
No other acquirer program on record names market data. A series built from licensed feeds raises the questions of the section on access licenses before it is offered, because the feed agreement may restrict derived data even where the firm did the calculation itself.
A firm unsure of that position can ask the vendor in writing before any buyer is approached.
Licensing Market Data for AI: Benefits, Risks, Downsides and Alternatives, Including Not Licensing
Licensing market data you own can earn from series already recorded, but data received under exchange or vendor agreements cannot be passed on without the agreement's permission, and a model trained on the data cannot be untrained.
Benefits
- A named ask exists: Mercor lists firm-owned time series and trading code.
- A firm keeps its own series and can license it without giving up the right to use it.
- Mercor states a whole-deal range for the listing, a claim and not a price.
Risks
- Breach of a vendor or exchange agreement if licensed data is passed on; a rights audit sorts own data from received data first.
- Declaration or audit duties tied to the declared use.
- Recorded series that embed licensed feeds.
- A reading of "non-display" that the exchange does not share, because the NYSE text has no AI wording.
Downsides
- No public price benchmark for market data is on record.
- Scoping work is needed to separate own, licensed and public data.
- A trained model keeps what it learned.
Alternatives
- Ask the vendor for a written AI training license.
- License only own-generated series or code.
- Keep the data for in-house use.
- Not licensing at all: the guide on whether to sell or license data sets out this outcome.
Each alternative leaves the exchange or vendor agreement as the first thing to read.
Neighboring Finance Data Types and the Seller Steps for Market Data
This page covers market and price data; a company's own ledgers, order and payment tables, sensor readings and the rules for financial customer records each have their own page.
- Financial records for AI training: ledgers and ERP records.
- Order and payment records: order and payment tables.
- Sensor and telemetry data: sensor readings.
- GLBA and financial customer data: financial customer records.
- Investment firms and hedge funds: the hedge fund listing.
- Data types AI companies buy: every data type.
The seller steps for market data you own follow the general process:
- Review data sale readiness before approaching a buyer.
- Use NDAs before sharing data with an AI buyer before any sample moves.
- Read AI data licensing agreements for the license clauses.
- Keep the exchange or vendor agreement beside the draft license.
Frequently Asked Questions
Can I train a model on exchange data I receive under a license?
Only as the agreement allows. NYSE's policy text has no AI wording and defines non-display use as accessing, processing or consuming data (document of 2022-03-21).
Does a non-display license cover AI training?
The NYSE text read does not say. The vendor or subscriber agreement controls, and a written answer from the vendor settles it.
Can a firm license its own price history to AI companies?
Yes, where the firm owns the series and holds the right to license it. Mercor asks for firm-owned series and lists redistributed feeds as not a fit.
How much do AI companies pay for market data?
No price per record is on record. Mercor states a whole-deal range of USD 25,000 to 1,150,000 (re-read 2026-11-17), a claim about a whole deal. Valuation methods are set out in what company data is worth to AI companies.