How much do AI companies pay for training data (14 figures)
The AI Training Data Price Index carries 14 published figures read from 14 sources' own pages, from 0.75 to 203,000,000 USD, read 31 October 2026.
- 91 to 33,833,334
- 333,833,334 to 67,666,667
- 167,666,667 to 101,500,000
- 101,500,000 to 135,333,334
- 135,333,334 to 169,166,667
- 1169,166,667 to 203,000,000
Bands
| Bucket | Value |
|---|---|
| 1 to 33,833,334 | 9 figures |
| 33,833,334 to 67,666,667 | 3 figures |
| 67,666,667 to 101,500,000 | 1 figure |
| 101,500,000 to 135,333,334 | 0 figures |
| 135,333,334 to 169,166,667 | 0 figures |
| 169,166,667 to 203,000,000 | 1 figure |
Across 33 sources read in full (31 October 2026), 14 publish a figure and 19 publish nothing.
By data type
| Breakdown | Median | Range | Sources |
|---|---|---|---|
| data typetext | 131,500,000USD | 60,000,000 to 203,000,000USD | 2 |
| data typevideo | 2.62USD | 0.75 to 6.25USD | 3 |
| data typeimage | 15USD | 15 to 15USD | 1 |
| data typebooks | 49,100,000USD | 5,000 to 75,000,000USD | 3 |
| data typenews | 20,000,000USD | 16,000,000 to 50,000,000USD | 3 |
| data typecode | 5,050USD | 100 to 10,000USD | 2 |
The figures, source by source
Below, every figure a source's page shows sits on its own line, linked and dated. Numbers that no source prints are left out.
| Source | Figure | Breakdown | Read |
|---|---|---|---|
| Reddit (data licensees, unnamed) - text data price | 203,000,000USD | text | 2026-10-29 |
| Informa / Taylor & Francis (AI licensees) - books data price | 75,000,000USD | books | 2026-10-31 |
| Google (Reddit deal) - text data price | 60,000,000USD | text | 2026-10-31 |
| Apple (publisher offers) - news data price | 50,000,000USD | news | 2026-10-31 |
| Wiley (AI licensees) - books data price | 49,100,000USD | books | 2026-10-31 |
| Amazon (New York Times deal) - news data price | 20,000,000USD | news | 2026-10-31 |
| OpenAI (Dotdash Meredith deal) - news data price | 16,000,000USD | news | 2026-10-31 |
| HUD AI (for frontier labs) - code data price | 10,000USD | code | 2026-10-31 |
| Microsoft (HarperCollins deal) - books data price | 5,000USD | books | 2026-10-31 |
| Gramian Consulting Group - code data price | 100USD | code | 2026-10-31 |
| Wirestock - image data price | 15USD | image | 2026-10-31 |
| Calliope Networks - video data price | 6.25USD | video | 2026-10-31 |
| Adobe - video data price | 2.62USD | video | 2026-10-31 |
| Troveo - video data price | 0.75USD | video | 2026-10-31 |
Where each figure is written (14 quotes)
- Reddit (data licensees, unnamed) - text data price, read 2026-10-29:
The aggregate transaction price of such arrangements totaled $203.0 million.
- Informa / Taylor & Francis (AI licensees) - books data price, read 2026-10-31:
Taylor & Francis had a strong year in 2025, with underlying revenue growth of 3.6% (excluding non-recurring data contracts) and further success in licencing, archives and data access, albeit non-recurring data access revenues were lower than the exceptional levels achieved in 2024 ($75m+).
- Google (Reddit deal) - text data price, read 2026-10-31:
The contract with Alphabet-owned Google is worth about $60 million per year, according to one of the sources.” FEB 22, 2024 content licensing, generative AI, Google, Reddit, Reuters
- Apple (publisher offers) - news data price, read 2026-10-31:
Proposed deals have been worth at least $50 million, and would allow Apple to license archives of news articles.
- Wiley (AI licensees) - books data price, read 2026-10-31:
AI license revenue was $49.1 million in the year ended April 30, 2026 compared to $40 million in the prior year.
- Amazon (New York Times deal) - news data price, read 2026-10-31:
Amazon's AI licensing deal with The New York Times is worth $20 million to $25 million per year, according to The Wall Street Journal.
- OpenAI (Dotdash Meredith deal) - news data price, read 2026-10-31:
OpenAI is paying the digital media company Dotdash Meredith at least $16 million per year to license its content, according to public financial documents reviewed by Adweek.
- HUD AI (for frontier labs) - code data price, read 2026-10-31:
Would you license your old/dead codebase as AI training data for $10k?
- Microsoft (HarperCollins deal) - books data price, read 2026-10-31:
The company has proposed a licensing agreement with publisher HarperCollins that would pay $5,000 per book for AI training rights.
- Gramian Consulting Group - code data price, read 2026-10-31:
Compensation For standard non-exclusive licensing, current reference rates are: $100 per 1M source-code tokens
- Wirestock - image data price, read 2026-10-31:
Payment for artists and photographers for licensing their work from AI training ranges from microtransactions of cents on the dollar when providers buy bulk data from Wirestock, up to $15-20 per artwork for specific needs, according to Khachatryan.
- Calliope Networks - video data price, read 2026-10-31:
The cost of licensing high-quality film and TV content can be significantly higher than licensing stock footage. For example, Calliope charges $6.25 per minute for HD content, with additional fees for 4K or 3D content.
- Adobe - video data price, read 2026-10-31:
Adobe is paying between 6 cents and 16 cents for each photo and an average of $2.62 per minute for videos, according to the report.
- Troveo - video data price, read 2026-10-31:
Troveo.ai offers upfront payments (typically $0.75–$3 per minute) for exclusive licensing rights to your owned video footage
Checked and silent on the figure
These 19 named sources were read on the dates shown and state no figure. They are part of the count and stay blank.
- Reddit (other revenue, incl. data licensing) - text data price, read 2026-10-31: every candidate on the page was refused
- Reddit (other revenue, incl. data licensing) - text data price, read 2026-10-29: every candidate on the page was refused
- Adobe - video data price, read 2026-10-31: read in full and no accepted figure on the page
- AI video-model buyers (via Bloomberg) - video data price, read 2026-10-08: every candidate on the page was refused
- Wiley (AI licensees) - books data price, read 2026-10-31: read in full and no accepted figure on the page
- Informa / Taylor & Francis (AI licensees) - books data price, read 2026-10-31: every candidate on the page was refused
- Wiley (AI licensees) - books data price, read 2026-10-31: every candidate on the page was refused
- Wiley (AI licensees) - books data price, read 2026-10-31: every candidate on the page was refused
- Shutterstock (Data, Distribution, and Services) - mixed data price, read 2026-10-31: every candidate on the page was refused
- Shutterstock (Data, Distribution, and Services) - mixed data price, read 2026-10-31: every candidate on the page was refused
- OpenAI (News Corp deal) - news data price, read 2026-10-31: read in full and no accepted figure on the page
- OpenAI (Axel Springer deal) - news data price, read 2026-10-31: read in full and no accepted figure on the page
- Perplexity (Comet Plus publishers) - news data price, read 2026-10-31: read in full and no accepted figure on the page
- Anthropic (Bartz settlement) - books data price, read 2026-10-31: read in full and no accepted figure on the page
- Newsflare - video data price, read 2026-10-31: read in full and no accepted figure on the page
- Neon Mobile - audio data price, read 2026-10-31: read in full and no accepted figure on the page
- Babel Audio - audio data price, read 2026-10-31: read in full and no accepted figure on the page
- Kled AI - image data price, read 2026-10-31: read in full and no accepted figure on the page
- Pangea (rates as quoted by aitrainer.work) - code data price, read 2026-10-31: read in full and no accepted figure on the page
The method in full
How it is built: we fetch each named page, keep the sentence carrying the figure verbatim, and pin it to a hash of the page as read.
Among 33 readable sources, 14 give a usable number and 19 give none; the silent ones stay listed and blank.
6 were unreadable and are not counted.
Medians, ranges and quartiles weigh every source equally, one figure each in USD.
A reviewer discarded 259 candidate sentence(s), with reasons on file.
The broader derivation this dataset is meant to support (not yet reflected above, which shows each source's own figure): one row per published price or deal value, with the buyer or platform, the data type, the unit (per hour, per minute, per image, per book, per million tokens, per codebase, per year, total) and the date it was published; across rows the median per unit and data type where three or more rows share one; a deal total is never divided into a unit price nobody published
259 candidate sentences on these pages failed review, and each failure is recorded with its reason.
Source list (39)
- Reddit (data licensees, unnamed) - text data price(quoted under the publisher's own terms)
- Reddit (other revenue, incl. data licensing) - text data price(quoted under the publisher's own terms)
- Reddit (other revenue, incl. data licensing) - text data price(quoted under the publisher's own terms)
- Google (Reddit deal) - text data price(quoted under the publisher's own terms)
- Adobe - video data price(quoted under the publisher's own terms)
- Adobe - video data price(quoted under the publisher's own terms)
- AI video-model buyers (via Bloomberg) - video data price(quoted under the publisher's own terms)
- Wirestock - image data price(quoted under the publisher's own terms)
- Defined.ai - image data price(quoted under the publisher's own terms)
- Microsoft (HarperCollins deal) - books data price(quoted under the publisher's own terms)
- Wiley (AI licensees) - books data price(quoted under the publisher's own terms)
- Informa / Taylor & Francis (AI licensees) - books data price(quoted under the publisher's own terms)
- Informa / Taylor & Francis (AI licensees) - books data price(quoted under the publisher's own terms)
- Wiley (AI licensees) - books data price(quoted under the publisher's own terms)
- Wiley (AI licensees) - books data price(quoted under the publisher's own terms)
- Wiley (AI licensees) - books data price(quoted under the publisher's own terms)
- Shutterstock (Data, Distribution, and Services) - mixed data price(quoted under the publisher's own terms)
- Shutterstock (Data, Distribution, and Services) - mixed data price(quoted under the publisher's own terms)
- Shutterstock (AI licensees) - mixed data price(quoted under the publisher's own terms)
- OpenAI (News Corp deal) - news data price(quoted under the publisher's own terms)
- OpenAI (Dotdash Meredith deal) - news data price(quoted under the publisher's own terms)
- Amazon (New York Times deal) - news data price(quoted under the publisher's own terms)
- OpenAI (Axel Springer deal) - news data price(quoted under the publisher's own terms)
- OpenAI (publisher offers) - news data price(quoted under the publisher's own terms)
- Apple (publisher offers) - news data price(quoted under the publisher's own terms)
- Perplexity (Comet Plus publishers) - news data price(quoted under the publisher's own terms)
- Anthropic (Bartz settlement) - books data price(quoted under the publisher's own terms)
- Calliope Networks - video data price(quoted under the publisher's own terms)
- Calliope Networks (founders' article) - video data price(quoted under the publisher's own terms)
- RecipeBook - video data price(quoted under the publisher's own terms)
- Troveo - video data price(quoted under the publisher's own terms)
- Newsflare - video data price(quoted under the publisher's own terms)
- Neon Mobile - audio data price(quoted under the publisher's own terms)
- Babel Audio - audio data price(quoted under the publisher's own terms)
- Kled AI - image data price(quoted under the publisher's own terms)
- Gramian Consulting Group - code data price(quoted under the publisher's own terms)
- HUD AI (for frontier labs) - code data price(quoted under the publisher's own terms)
- Pangea (rates as quoted by aitrainer.work) - code data price(quoted under the publisher's own terms)
- sec.gov(quoted under the publisher's own terms)
Reference or embed these figures
The AI Training Data Price Index carries 14 published figures read from 14 sources' own pages, from 0.75 to 203,000,000 USD, read 31 October 2026.
Across 33 sources read in full (31 October 2026), 14 publish a figure and 19 publish nothing.
AI Training Data Price Index, data type text: median 131,500,000 USD across 2 sources, read 31 October 2026.
AI Training Data Price Index, data type video: median 2.62 USD across 3 sources, read 31 October 2026.
AI Training Data Price Index, data type image: median 15 USD across 1 source, read 31 October 2026.
AI Training Data Price Index, data type books: median 49,100,000 USD across 3 sources, read 31 October 2026.
AI Training Data Price Index, data type news: median 20,000,000 USD across 3 sources, read 31 October 2026.
AI Training Data Price Index, data type code: median 5,050 USD across 2 sources, read 31 October 2026.
Credit as: "datasetvaluation.com AI Training Data Price Index", updated 2026-10-31, https://datasetvaluation.com/datasets/ai-training-data-price-index/.
Get the CSV (14 rows as of 2026-10-31)
Questions this data answers
What have AI buyers paid for training data, by data type and unit: per hour of video or audio, per image, per book, per token, per codebase, and in published licensing deals, in US dollars?
The AI Training Data Price Index carries 14 published figures read from 14 sources' own pages, from 0.75 to 203,000,000 USD, read 31 October 2026.
What is the published figure for video in the AI Training Data Price Index?
For video, the median published figure is $2.62 across 3 figures from the sources' own pages, from $0.75 to $6.25, read 31 Oct 2026.
What is the published figure for books in the AI Training Data Price Index?
For books, the median published figure is $49,100,000 across 3 figures from the sources' own pages, from $5,000 to $75,000,000, read 31 Oct 2026.
What is the published figure for news in the AI Training Data Price Index?
For news, the median published figure is $20,000,000 across 3 figures from the sources' own pages, from $16,000,000 to $50,000,000, read 31 Oct 2026.
Where do the figures in the AI Training Data Price Index come from?
Every figure is quoted from its source's own page and dated (last read 31 Oct 2026). 33 named sources were read and 14 publish a figure; the 19 that publish none are recorded as not published, never estimated.