AI data licensing deals, tracked

Who has paid for data to train or power AI, how much and on what terms. Every deal comes from public reporting and links to a dated source, and the whole list downloads as a CSV.

Explore the deals Download the CSV

This tracker lists 100 public AI data deals from January 2023 to September 2026, each with the buyer, the seller, the data type, the deal type, the disclosed value and a dated source. Examples include Google's winning $10M bid for bankrupt Spirit Airlines' archive, pending court approval, and HarperCollins' backlist license at $5,000 a title for 3 years.

Google's $10M bid won the bankruptcy auction for Spirit Airlines' archive. TIME, 25 Aug 2026 (opens in a new tab)
The sale still needs court approval, so the bid stays out of every total until a court approves it. Business Insider, 5 Oct 2026 (opens in a new tab)
News Corp's deal with OpenAI was reported at more than $250M over five years. Variety, 22 May 2024 (opens in a new tab)
HarperCollins' backlist license with Microsoft pays $5,000 per title for 3 years. Publishers Weekly, 19 Nov 2024 (opens in a new tab)
OpenAI is the buyer in 30 of the 100 deals, more than any other company. Rightmark Deal Tracker, 10 Oct 2026

Updated 10 Oct 2026

Deals: 100; Each with a dated source
Years covered: 2023 to 2026; January 2023 to September 2026
Disclosed value, licenses and sales: $1.01B; Across the 23 licenses and sales with a US dollar figure: a floor that mixes one-off, annual and multi-year values as reported. Acquisitions, $47.4B, are counted apart, and a winning bid pending court approval is left out.
Most active buyer: OpenAI; 30 of 100 deals. Google is next, with 10.

Explore the deals

Deals per quarter, by deal type

Training
Training and live access
Live access
Sale or acquisition
Winning bid, pending approval

Chart data as a table

Deals per quarter by deal type
Quarter	Training	Training and live access	Live access	Sale	Acquisition	Winning bid	All deals
Q1 2023Jan to Mar 2023	1	0	0	0	0	0	1
Q2 2023Apr to Jun 2023	0	0	0	0	0	0	0
Q3 2023Jul to Sep 2023	3	0	0	0	0	0	3
Q4 2023Oct to Dec 2023	2	1	0	0	0	0	3
Q1 2024Jan to Mar 2024	1	3	1	0	0	0	5
Q2 2024Apr to Jun 2024	8	7	1	0	0	0	16
Q3 2024Jul to Sep 2024	2	1	3	0	0	0	6
Q4 2024Oct to Dec 2024	1	0	4	0	0	0	5
Q1 2025Jan to Mar 2025	1	0	6	0	1	0	8
Q2 2025Apr to Jun 2025	2	1	3	0	1	0	7
Q3 2025Jul to Sep 2025	2	0	3	0	1	0	6
Q4 2025Oct to Dec 2025	5	0	5	0	0	0	10
Q1 2026Jan to Mar 2026	1	2	3	0	1	0	7
Q2 2026Apr to Jun 2026	3	0	6	2	0	0	11
Q3 2026Jul to Sep 2026	4	0	7	0	0	1	12
All quarters	36	15	42	2	4	1	100

Showing 21 of 100 deals, newest first

AI data licensing deals, January 2023 to September 2026. Each row has its notes below it. Source links open in a new tab.
Date	Buyer	Seller or source	Data	Deal type	Disclosed value	Term	Sources
Sep 2026	Undisclosed precision medicine AI company	OneMedNetOther	OtherHealth: clinical and imaging data	Training	Seven-figure multiyear agreementDisclosed	Multi-year	OneMedNet via GlobeNewswire, 30 Sep 2026 StockTitan, 30 Sep 2026
Notes on Sep 2026, Undisclosed precision medicine AI company and OneMedNet OneMedNet supplies de-identified clinical and imaging data from more than 2,300 healthcare sites to train clinical AI models and build derivative datasets. OneMedNet via GlobeNewswire, 30 Sep 2026 Copy a link to this deal
Sep 2026	Arcee AI	Bolt.new Pro users (opt-in, via Bolt Forge)Creator program	Code	Training	50 times more open-model usage for contributors; no cash payment reported	Research preview to Oct 2026	Yahoo Finance, 14 Sep 2026 IT Brief, 17 Sep 2026
Notes on Sep 2026, Arcee AI and Bolt.new Pro users (opt-in, via Bolt Forge) Developers who opt in share anonymized coding sessions to train open-weight models that are released publicly. Yahoo Finance, 14 Sep 2026 Copy a link to this deal
Sep 2026	Undisclosed voice AI company	SilencioMarketplace	Audio	Training	$250,000 contractDisclosed	Not reported	Advanced Blockchain AG via FinanzWire, 7 Sep 2026
Notes on Sep 2026, Undisclosed voice AI company and Silencio Silencio supplies single and multi-speaker speech recordings in Tagalog and Cebuano, its largest contract to date. Advanced Blockchain AG via FinanzWire, 7 Sep 2026 Copy a link to this deal
Sep 2026	Google	About 100 publishers and websitesPublisher	OtherWeb and news content	Live access	Usage-based; reported payments range from under $1,000 to more than $1M a year	Not reported	MediaNama, 17 Sep 2026 PPC Land, 30 Sep 2026
Notes on Sep 2026, Google and About 100 publishers and websites Google pays sites whose content significantly shapes AI Overviews, AI Mode and Gemini answers, through Search Console. MediaNama, 17 Sep 2026 Copy a link to this deal
Sep 2026	OpenAI	Indian Express GroupPublisher	News	Live access	About $3M a yearEstimate from industry sources	Not reported	exchange4media, 23 Sep 2026 MediaBrief, 8 Sep 2026
Notes on Sep 2026, OpenAI and Indian Express Group Indian Express journalism appears in ChatGPT with attribution and links. exchange4media, 23 Sep 2026 Copy a link to this deal
Sep 2026	OpenAI	Bennett, Coleman & Co (The Times of India)Publisher	News	Live access	About $5M a yearEstimate from industry sources	Not reported	exchange4media, 23 Sep 2026 MediaNama, 8 Sep 2026
Notes on Sep 2026, OpenAI and Bennett, Coleman & Co (The Times of India) ChatGPT shows excerpts and summaries of journalism from The Times of India and The Economic Times with attribution and links. exchange4media, 23 Sep 2026 Copy a link to this deal
Sep 2026	OpenAI	Village MediaPublisher	News	Live access	Mix of funding, API credits and technical supportReported	Not reported	Axios, 15 Sep 2026 Media in Canada, 21 Sep 2026
Notes on Sep 2026, OpenAI and Village Media Village Media content is cited in ChatGPT, and the deal funds a local services tool called Open Door. Axios, 15 Sep 2026 Copy a link to this deal
Aug 2026	OpenEvidence	Springer NaturePublisher	Academic	Live access	Not disclosed	Not reported	Springer Nature, 4 Aug 2026
Notes on Aug 2026, OpenEvidence and Springer Nature Physicians on OpenEvidence can retrieve and cite Springer Nature's peer-reviewed medical content. Springer Nature, 4 Aug 2026 Copy a link to this deal
Aug 2026	Figure AI	Contributors on Figure's Index appCreator program	Video	Training	$15M paid to contributors so farReported	Not reported	Forbes via Yahoo Finance, 26 Aug 2026
Notes on Aug 2026, Figure AI and Contributors on Figure's Index app Contributors wear cameras and sensors while doing household and commercial tasks to produce robot training data. Forbes via Yahoo Finance, 26 Aug 2026 In the Rightmark Price IndexFirst-person task video$15 to $50 per hour Copy a link to this deal
Aug 2026	Google	Spirit AirlinesCompany archive	Company records	Winning bidPending court approval	$10MWinning bankruptcy auction bid, not in totals	Not reported	TIME, 25 Aug 2026 Business Insider, 5 Oct 2026 The Next Web, 18 Sep 2026
Notes on Aug 2026, Google and Spirit Airlines Google's $10M bid won the bankruptcy auction for an archive of about 100M emails, 500M Teams messages and 30M lines of code. The sale still needs court approval: the hearing is set for 14 October 2026, the court's privacy ombudsman recommended approval on 5 October, and a rival bidder offered $12.5M after the auction. TIME, 25 Aug 2026 Totals Left out of every total until a court approves the sale. In the Rightmark Price IndexLarge enterprise archive$10M per company Copy a link to this deal
Jul 2026	OpenAI	YelpPlatform	OtherReviews and business listings	Live access	Not disclosed	Not reported	Axios, 23 Jul 2026 citybiz, 23 Jul 2026
Notes on Jul 2026, OpenAI and Yelp ChatGPT uses Yelp reviews, photos and business information for local recommendations, with Yelp branding and links. Axios, 23 Jul 2026 Copy a link to this deal
Jul 2026	Microsoft	Nine EntertainmentPublisher	News	Live access	Not disclosed	Not reported	Microsoft, 3 Jul 2026 Cyber Daily, 6 Jul 2026
Notes on Jul 2026, Microsoft and Nine Entertainment Copilot can reference the full text of mastheads such as The Sydney Morning Herald and The Age to ground its answers. Microsoft, 3 Jul 2026 Copy a link to this deal
Jun 2026	OpenAI	Getty ImagesMarketplace	Images	Live access	Not disclosed	Multi-year	PetaPixel, 22 Jun 2026 The Decoder, 22 Jun 2026
Notes on Jun 2026, OpenAI and Getty Images ChatGPT search shows licensed Getty images, and the announcement did not mention training. PetaPixel, 22 Jun 2026 Copy a link to this deal
Jun 2026	Google	Android app developers (pilot)Creator program	Code	Training	Not disclosed	Not reported	404 Media, 3 Jun 2026
Notes on Jun 2026, Google and Android app developers (pilot) Google invited Play Store developers to be paid for sharing the source code of live and archived apps. 404 Media, 3 Jun 2026 Copy a link to this deal
Jun 2026	Udio	NMPA member music publishers (opt-in)Publisher	Music	Training	Not disclosed	Not reported	Music Business Worldwide, 11 Jun 2026
Notes on Jun 2026, Udio and NMPA member music publishers (opt-in) An industry-wide opt-in license lets member publishers' songs be used to train Udio's models. Music Business Worldwide, 11 Jun 2026 Copy a link to this deal
Jun 2026	OpenAI	Associated PressPublisher	News	Live access	Not disclosed	Through the 2028 US election season	MediaPost, 1 Jun 2026
Notes on Jun 2026, OpenAI and Associated Press ChatGPT will use AP's vote count data for elections in the US and Brazil. MediaPost, 1 Jun 2026 Copy a link to this deal
May 2026	Spotify	Universal Music GroupPublisher	Music	Live access	Not disclosed	Not reported	Music Business Worldwide, 21 May 2026
Notes on May 2026, Spotify and Universal Music Group Spotify Premium users will be able to make AI covers and remixes of songs whose artists and songwriters opt in. Music Business Worldwide, 21 May 2026 Copy a link to this deal
May 2026	OpenAI	Grupo Folha and UOLPublisher	News	Live access	Not disclosed	Not reported	LatAm Journalism Review, 26 May 2026
Notes on May 2026, OpenAI and Grupo Folha and UOL ChatGPT shows summaries of the outlets' reporting with links, and the agreement ended Folha's lawsuit against OpenAI. LatAm Journalism Review, 26 May 2026 Copy a link to this deal
May 2026	Snowflake customers (via Snowflake Cortex)	AP, People Inc., USA Today Network, The Washington Post and 13 other publishersPublisher	News	Live access	Six-figure deals, flat fee or usage-basedReported	Not reported	Digiday, 29 May 2026
Notes on May 2026, Snowflake customers (via Snowflake Cortex) and AP, People Inc., USA Today Network, The Washington Post and 13 other publishers Enterprises on Snowflake license publisher content for retrieval-based AI, and Snowflake takes no share of the fees. Digiday, 29 May 2026 Copy a link to this deal
Apr 2026	Undisclosed AI data buyers	Nearly 100 shut-down startups (via SimpleClosure)Company archive	Company records	Sale	More than $1M in total; $10,000 to $100,000 per companyReported	Not reported	Forbes, 16 Apr 2026 Gizmodo, 17 Apr 2026
Notes on Apr 2026, Undisclosed AI data buyers and Nearly 100 shut-down startups (via SimpleClosure) SimpleClosure brokered sales of source code and workplace data from closing startups over about a year. Forbes, 16 Apr 2026 In the Rightmark Price IndexShut-down startup, code and workspace$10,000 to $100,000 per company Copy a link to this deal
Jan 2023	Meta	ShutterstockMarketplace	Images	Training	$25M to $50MReported range for each of Shutterstock's big tech deals	Not reported	Shutterstock via PR Newswire, 12 Jan 2023 Reuters via Rappler, 7 Apr 2024
Notes on Jan 2023, Meta and Shutterstock Shutterstock expanded its relationship with Meta to supply licensed images and metadata for Meta's generative AI work. Shutterstock via PR Newswire, 12 Jan 2023 Copy a link to this deal

Show all 100 deals

From these deals to your data

Each data type has its own page on what buyers pay, what sets the price and what blocks a deal. To see what buyers are sourcing right now, read Wanted now.

How we build the tracker

100 deals, each checked against a party's statement or an independent report. The same rules apply to every row.

What counts: A license or sale of content or data for AI training or AI products, a creator footage program, a bankruptcy or shutdown data sale, a data marketplace transaction, or the purchase of a data supplier or a stake in one. A winning bankruptcy auction bid is listed with its status while it waits for court approval.
Public facts only: A named publication or one of the parties reported who bought or licensed the data, who supplied it, what it covers and when. Nothing here comes from Rightmark's own work for owners, and we never publish terms that were not reported publicly.
Dates: The month a deal was announced or first reported, which can be later than signing. One announcement is one row, even when it covers several publishers or several buyers.
Deal types: Training: a reported right to train models. Training and live access: training rights plus use in AI answers or generated output. Live access: content used in AI answers or generated output, with no reported training right. Sale: an outright purchase of a dataset. Acquisition: a purchase of a data supplier or an equity stake in one. Winning bid: a winning bid in a bankruptcy auction that still needs court approval. It is not a sale until a court approves it, so it stays out of every total and of the sales count.
Values: Shown as reported, with estimates labeled. For sorting and totals each deal carries one US dollar figure: the lower end of a range, the stated maximum of an 'up to' figure and one year of an annual figure. Unit prices, euro amounts and undisclosed terms stay out of totals, so totals mix one-off, annual and multi-year values and are a floor, not the size of the market. Totals add licenses and sales only. Acquisitions are counted apart, and a winning bid that still needs court approval stays out of every total and of the sales count until a court approves the sale.
Left out: Settlements for past use, negotiations and talks, losing bids, funding rounds, offers to authors with no named buyer, partnerships with no reported payment for data and synthetic data companies.
Sources: Every row links to at least one dated source. In our last check, completed on 10 Oct 2026, a source loaded for 99 of 100 rows. We started from the public deal counts published by Media and the Machine and the deal lists kept by Digiday, Press Gazette and Ithaka S+R, then checked each deal against a party's statement or an independent report.
Corrections: Spotted an error or a missing deal? Email hello@rightmark.ai with a link to the source.

Cite and reuse

Free to reuse under CC BY 4.0 (opens in a new tab), with attribution to the Rightmark Deal Tracker and a link to this page.

Rightmark Deal Tracker: AI data licensing deals, January 2023 to September 2026. Updated 10 Oct 2026. https://rightmark.ai/deal-tracker/

Download CSV Download JSON

The CSV and JSON hold every column on this page plus the source links, the value we use for sorting and totals, and whether each deal counts in the totals.

Questions

See the Rightmark Price Index for what buyers pay per unit, by data type.

Which AI companies have signed data licensing deals?

The buyers named in the most deals here are OpenAI, Google, Meta, Microsoft and Perplexity. OpenAI is the buyer in the most deals in this tracker, 30 of 100. Each deal is listed with its date and a source.

How much do AI data deals pay?

Reported unit prices include unused creator footage from $1 a minute and nonfiction backlist titles at $5,000 a title for 3 years. The largest disclosed deal value here is more than $250M over five years for News Corp's deal with OpenAI. A winning bid is not a sale: Google's $10M bid for bankrupt Spirit Airlines' archive is pending court approval. Typical prices for company archives are in the Price Index.

Sources: The Decoder, 11 Jan 2025 (opens in a new tab); eWeek, 7 Dec 2024 (opens in a new tab); Variety, 22 May 2024 (opens in a new tab); TIME, 25 Aug 2026 (opens in a new tab)

Why do the totals mix annual and one-off figures?

Because deals are reported that way. We add each published US dollar figure as reported: the lower end of a range, the maximum of an 'up to' figure and one year of an annual figure. So the $1.01B total for licenses and sales is a floor for the public market, not its size. Acquisitions are counted apart, and a winning bid that still needs court approval stays out of every total until the sale is approved.

Do you include rumored or private deals?

No. We list only deals reported by a named publication or disclosed by a party, with the date and a link. Negotiations, losing bids, settlements and funding rounds are left out, and we never publish terms that were not reported publicly.

How often is the tracker updated?

The date at the top of the page shows the last update. We add a deal only after checking it against a named source. To cite a fixed version, quote that date or keep a copy of the CSV.

Can I use the tracker data?

Yes. It is free to reuse under CC BY 4.0, with attribution to the Rightmark Deal Tracker and a link. Download the CSV or JSON, or email hello@rightmark.ai about a deal we missed.

Hold data like this? See what it is worth.

About a minute, no files, no obligation. You will see a range, the routes it fits and what would raise the price.

Appraise your data

How the Deal Tracker classifies public records

The Deal Tracker is a record of public reports about data transactions. It does not show current demand or a relationship that is not in the cited record.

The tracker should use the term supported by the source. Do not upgrade a training arrangement to a sale, or a bid to a completed transaction.

Keep the primary public link and the date used for the row. A source check should confirm who reported what and when. Do not add a row for a relationship merely because it sounds commercially plausible.

Research note template

Record: [short source-backed description]
Public source: [URL]
Source date: [date or month]
Deal type: [one current type]
Value text: [verbatim source description or undisclosed]
Value treatment: [reported, range, up to, annual, estimate, excluded]
Status: [completed or pending]
Notes: [unit, scope, and source limitation]

The record can retain a counterparty name when the public source does. The surrounding copy should not add a buyer, seller, partner, demand signal, or relationship beyond the record.

Data reuse

General information, not legal advice.

When reusing public records, preserve their source links, dates, limitations and applicable attribution requirements. Do not infer a new price or market conclusion from a row without a current source.

Use the appraisal path for a separate review of a specific collection. A tracker row is not an appraisal.

Public deal calendar and filter guide

Use the filters to narrow the records that are already in the source dataset. Treat disclosed, estimated, reported, and undisclosed values differently: a missing value is a record state, not a zero.

type: a deal type from the filter menu. Read the row definition, not only the label.
year: the calendar year of the record date.
data: a data type from the filter menu, not a Price Index family. Do not infer it from a headline.
buyer: the source-owned buyer field where present, not a relationship or endorsement.
quarter: a grouping of dated records, not a market statistic.
q: a cleaned, bounded text search. Check the returned row and its source link.
sort: a display order. Unknown values stay outside numeric totals.
open: the record to open. Open its source before repeating a claim.

Filter values the page does not recognize are ignored. Keep the URL and the returned row together when citing a result. Read the returned row as a source record, not as a new dataset created by the filter.

Open a record in the deal table and read its source before using it in research.