Home / Research / AI Training Data Licensing Deals

Market research · October 6, 2026

AI Training Data Licensing Deal Statistics: 2026 Report

AI developers that once collected training data from the open web now increasingly pay for it, through licensing agreements with publishers, platforms, image libraries, and data brokers. We counted 312 of those agreements signed between January 2023 and September 2026, drawing on company announcements, regulatory filings, court records, and confirmed press reports. Quarterly deal volume rose from 3 in Q1 2023 to 39 in Q3 2026, while the median annual value of a deal fell from $9.4M in Q2 2023 to $3.6M.

Deal values were disclosed or reliably reported for 147 of the 312 agreements. Every dollar figure in this report comes from those 147 deals and is expressed as an annual value.

AI Training Data Licensing Deals by Licensor Type

The table below shows how many deals each type of content owner signed, the median annual value of those deals, and how often the agreement also granted rights to display the content in AI answers as well as train on it.

AI Training Data Licensing Deals by Licensor Type, 2026

Licensor TypeDealsDeals With Disclosed ValueMedian Annual ValueMedian Term (Years)Share Including Display Rights
News publishers10447$4.2M2.058.7%
Specialized datasets and data brokers5219$2.4M1.519.2%
Book and academic publishers4722$6.8M3.08.5%
Stock image and video libraries4124$11.5M3.012.2%
Individual creators and creator platforms279$1.1M1.07.4%
Music labels and catalogs2313$9.6M3.54.3%
Social and forum platforms1813$38.0M2.566.7%
All licensors312147$4.6M2.030.5%
Horizontal bar chart of median annual deal value in millions of dollars by licensor type, from $38M for social and forum platforms down to $1.1M for individual creators and creator platforms.
Median annual deal value by licensor type ($M), 2026

Three findings stood out to our team in this data:

AI Training Data Licensing Deals by Quarter, Q1 2023 to Q3 2026

The pace of dealmaking rose in almost every quarter we tracked. The table below shows the number of deals signed each quarter and the median annual value of those with disclosed terms, followed by a graph of quarterly and cumulative deal counts.

AI Training Data Licensing Deals by Quarter, 2026

QuarterDeals SignedMedian Annual Value of Disclosed Deals
Q1 20233Not disclosed
Q2 20235$9.4M
Q3 20237$7.8M
Q4 20239$8.1M
Q1 202414$6.9M
Q2 202417$7.2M
Q3 202419$6.1M
Q4 202422$5.8M
Q1 202524$5.2M
Q2 202526$4.9M
Q3 202523$5.1M
Q4 202531$4.4M
Q1 202635$4.1M
Q2 202638$3.8M
Q3 202639$3.6M
Bar and line chart of AI training data licensing deals signed in each quarter from Q1 2023 to Q3 2026, with a line for the cumulative total on a second axis.
AI training data licensing deals per quarter and cumulative total, Q1 2023 to Q3 2026

The quarterly data led our team to three conclusions:

AI Training Data Licensing Deals by Buyer Type

We grouped buyers into five categories. The table below shows how many deals each category signed and what it paid.

AI Training Data Licensing Deals by Buyer Type, 2026

Buyer TypeDealsShare of All DealsDeals With Disclosed ValueMedian Annual Value
Frontier model developers13142.0%64$7.1M
Large technology platforms7825.0%41$8.4M
AI search and answer engines4313.8%17$1.9M
Vertical and enterprise AI developers3912.5%16$1.6M
Data marketplaces and intermediaries216.7%9$2.2M

Frontier model developers, the companies building the largest general-purpose AI models, and large technology platforms together signed 209 deals, 67.0% of the total, and paid the highest median annual values at $7.1M and $8.4M. AI search and answer engines signed 43 deals at a much lower median of $1.9M, and their agreements were weighted toward news content and display rights, since these products need current information that they can quote and link. Data marketplaces and intermediaries, which license content from many small owners and resell it as packaged datasets, accounted for 21 deals and are the fastest-growing buyer category in the most recent quarters of our data.

How AI Training Data Licensing Deals Are Structured

The way buyers pay for data has shifted over the period. The table below compares the primary payment structure of deals signed in 2023 with those signed in 2026 through September.

Payment Structures of AI Training Data Licensing Deals, 2026

Primary Payment StructureShare of 2023 Deals (24 Deals)Share of 2026 Deals (112 Deals)Change (Points)
Flat annual fee58.3%34.8%-23.5
Usage-based or per-unit fee12.5%26.8%+14.3
Revenue share4.2%13.4%+9.2
Flat fee plus usage component16.7%18.7%+2.0
Technology, credits, or equity in lieu of cash8.3%6.3%-2.0
Slope chart of the share of deals using each primary payment structure in 2023 and 2026: flat annual fee, usage-based or per-unit fee, revenue share, flat fee plus usage component, and technology, credits, or equity in lieu of cash.
Shift in payment structure, 2023 to 2026 (% of deals)

Three patterns in the deal structure data stood out to us:

AI Training Data Prices per Unit

A subset of deals, mostly with image libraries, video sources, audio providers, and book publishers, priced content per unit. The table below gives the median of those unit prices and the interquartile range, which runs from the 25th percentile price to the 75th percentile price.

AI Training Data Prices per Unit, 2026

Content TypePricing UnitDeals With Unit PricingMedian PriceInterquartile Range
Still imagePer image38$1.12$0.41 to $3.48
Video footagePer minute29$2.65$1.05 to $6.20
Transcribed speech audioPer hour17$58$22 to $141
BookPer title21$4,300$1,850 to $9,600
News article archivePer article14$0.46$0.18 to $1.30

Unit prices at the 75th percentile were five to nine times those at the 25th percentile for every content type. Three characteristics explained most of that range in the deals we examined: exclusivity, the presence of rich metadata such as captions and rights information, and whether the content was already publicly accessible online. Video footage commanded the strongest price growth over the period, as developers of video generation models competed for licensed, high-resolution footage with clear rights.

Sources