AI training data statistics 2027
Every statistic on this page is one sentence you can cite, with its source, its date and a link, grouped by topic and checked in November 2026.
By Gareth HoyleLast updated How we research
In short
The AI training data market was worth $3.2 billion to $8.74 billion in 2025, according to estimates from Grand View Research and Mordor Intelligence. The largest reported license is News Corp's, at about $250 million over five years from OpenAI, and the largest settlement is $1.5 billion in Bartz v. Anthropic. Epoch AI expects public human text to be used up between 2026 and 2032.
Key takeaways
- $1.5 billion, about $3,000 per book, is what Anthropic agreed to pay in Bartz v. Anthropic, and the court gave the settlement final approval on July 20, 2026, according to the Association of American Publishers.
- About $250 million over five years in cash and credits is the reported value of News Corp's content license to OpenAI, the largest value in our deals tracker, as reported by The Wall Street Journal in May 2024.
- 2026 to 2032 is the window in which language models trained on current trends will have used the full stock of public human text, according to Epoch AI's estimate of June 6, 2024 (80% confidence interval).
- 45% of C4 is restricted by terms of service crawling restrictions, the Data Provenance Initiative reported in July 2024.
- More than $1.5 million a day is paid out by Mercor to a roster of more than 30,000 experts, TechCrunch reported on October 27, 2025.
How big is the AI training data market?
The AI training data market was worth $3.2 billion to $8.74 billion in 2025, according to research firm estimates that each define the market differently. US company filings give narrower but harder numbers for individual sellers such as Reddit and Shutterstock.
$3.2 billion was Grand View Research's estimate of the AI training dataset market in 2025, with a forecast of $16.3 billion by 2033 at 22.6% a year, in its report published in June 2026.
EstimateSource: Grand View Research, June 2026On our site: How big is the AI training data market?Link
Cite this statistic
$3.2 billion was Grand View Research's estimate of the AI training dataset market in 2025, with a forecast of $16.3 billion by 2033 at 22.6% a year, in its report published in June 2026. Source: Grand View Research. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#market-grand-view-2025
$8.74 billion was Mordor Intelligence's estimate of the AI training dataset market in 2025, with a forecast of $49.82 billion by 2031 at 33.14% a year, in its report updated on May 26, 2026.
EstimateSource: Mordor Intelligence, May 26, 2026On our site: How big is the AI training data market?Link
Cite this statistic
$8.74 billion was Mordor Intelligence's estimate of the AI training dataset market in 2025, with a forecast of $49.82 billion by 2031 at 33.14% a year, in its report updated on May 26, 2026. Source: Mordor Intelligence. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#market-mordor-2025
$2.82 billion in 2024, growing to $9.58 billion by 2029 at 27.7% a year, is the AI training dataset market estimate MarketsandMarkets published on November 26, 2024.
EstimateSource: MarketsandMarkets via GlobeNewswire, November 26, 2024On our site: How big is the AI training data market?Link
Cite this statistic
$2.82 billion in 2024, growing to $9.58 billion by 2029 at 27.7% a year, is the AI training dataset market estimate MarketsandMarkets published on November 26, 2024. Source: MarketsandMarkets via GlobeNewswire. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#market-marketsandmarkets-2024
$203.0 million was the aggregate contract value of the data licensing arrangements Reddit entered in January 2024, with terms of two to three years, according to its Form S-1 filed with the SEC in February 2024.
Source: US Securities and Exchange Commission (EDGAR), February 2024On our site: How big is the AI training data market?Link
Cite this statistic
$203.0 million was the aggregate contract value of the data licensing arrangements Reddit entered in January 2024, with terms of two to three years, according to its Form S-1 filed with the SEC in February 2024. Source: US Securities and Exchange Commission (EDGAR). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#market-reddit-s1
$114.7 million of other revenue in 2024, up from $15.2 million in 2023, is reported in Reddit's Form 10-K for 2024, which says other revenue grew as a result of content licensing agreements executed in 2024 (the line also includes Reddit Premium and user economy products).
Source: US Securities and Exchange Commission (EDGAR), February 2025On our site: How big is the AI training data market?Link
Cite this statistic
$114.7 million of other revenue in 2024, up from $15.2 million in 2023, is reported in Reddit's Form 10-K for 2024, which says other revenue grew as a result of content licensing agreements executed in 2024 (the line also includes Reddit Premium and user economy products). Source: US Securities and Exchange Commission (EDGAR). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#market-reddit-10k-2024
$137.3 million of Data, Distribution, and Services revenue in 2023, up 256%, is reported in Shutterstock's Form 10-K for 2023, with its data offering accounting for $84.9 million of the growth.
Source: US Securities and Exchange Commission (EDGAR), 2023On our site: How big is the AI training data market?Link
Cite this statistic
$137.3 million of Data, Distribution, and Services revenue in 2023, up 256%, is reported in Shutterstock's Form 10-K for 2023, with its data offering accounting for $84.9 million of the growth. Source: US Securities and Exchange Commission (EDGAR). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#market-shutterstock-10k-2023
$75 million or more of non-recurring data access revenue, earned in particular from AI companies, was included in Taylor and Francis's 12.8% revenue growth in 2024, according to Informa's full-year results published on March 6, 2025.
Source: Informa PLC via the FCA National Storage Mechanism, March 6, 2025On our site: Selling books and publishing archives to AI companiesLink
Cite this statistic
$75 million or more of non-recurring data access revenue, earned in particular from AI companies, was included in Taylor and Francis's 12.8% revenue growth in 2024, according to Informa's full-year results published on March 6, 2025. Source: Informa PLC via the FCA National Storage Mechanism. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#market-taylor-francis-2024
Over $1 billion in 2024 revenue made Surge AI larger than Scale AI, which had $870 million, Reuters reported on July 1, 2025, citing sources.
Source: Reuters via The Star, July 1, 2025On our site: Surge AILink
Cite this statistic
Over $1 billion in 2024 revenue made Surge AI larger than Scale AI, which had $870 million, Reuters reported on July 1, 2025, citing sources. Source: Reuters via The Star. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#market-surge-scale-revenue
What have AI data licensing deals been worth?
AI data licensing deals with a reported value range from per-minute video payments to about $250 million over five years, but 14 of the 25 deals we track were announced without terms.
About $250 million over five years in cash and credits is the reported value of News Corp's content license to OpenAI, the largest value in our deals tracker, as reported by The Wall Street Journal in May 2024.
Source: Engadget, citing The Wall Street Journal, May 2024On our site: AI data deals trackerLink
Cite this statistic
About $250 million over five years in cash and credits is the reported value of News Corp's content license to OpenAI, the largest value in our deals tracker, as reported by The Wall Street Journal in May 2024. Source: Engadget, citing The Wall Street Journal. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#deals-news-corp-openai
About $60 million a year is the reported value of Google's license to train AI models on Reddit content, Reuters reported on February 22, 2024, citing one source.
Source: Reuters via Nieman Lab, February 22, 2024On our site: AI data deals trackerLink
Cite this statistic
About $60 million a year is the reported value of Google's license to train AI models on Reddit content, Reuters reported on February 22, 2024, citing one source. Source: Reuters via Nieman Lab. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#deals-reddit-google
$20 million to $25 million a year is the reported value of Amazon's license for New York Times editorial content, which covers AI model training, according to The Wall Street Journal in July 2025.
Source: Engadget, citing The Wall Street Journal, July 2025On our site: AI data deals trackerLink
Cite this statistic
$20 million to $25 million a year is the reported value of Amazon's license for New York Times editorial content, which covers AI model training, according to The Wall Street Journal in July 2025. Source: Engadget, citing The Wall Street Journal. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#deals-nyt-amazon
At least $16 million a year is the fixed payment OpenAI makes to Dotdash Meredith to license its content, Adweek reported in November 2024 from public financial documents.
Source: Engadget, citing Adweek, November 2024On our site: AI data deals trackerLink
Cite this statistic
At least $16 million a year is the fixed payment OpenAI makes to Dotdash Meredith to license its content, Adweek reported in November 2024 from public financial documents. Source: Engadget, citing Adweek. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#deals-dotdash-openai
$5,000 per title, split equally between author and publisher, is the reported fee for HarperCollins nonfiction backlist titles licensed for AI training over three years, Publishers Weekly reported in November 2024.
Source: Publishers Weekly, November 2024On our site: Selling books and publishing archives to AI companiesLink
Cite this statistic
$5,000 per title, split equally between author and publisher, is the reported fee for HarperCollins nonfiction backlist titles licensed for AI training over three years, Publishers Weekly reported in November 2024. Source: Publishers Weekly. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#deals-harpercollins-per-title
$25 million to $50 million is the likely value of Shutterstock's license of millions of images to Apple for AI training, Reuters reported in April 2024, and Apple has not confirmed the deal.
Source: 9to5Mac, citing Reuters, April 2024On our site: AI data deals trackerLink
Cite this statistic
$25 million to $50 million is the likely value of Shutterstock's license of millions of images to Apple for AI training, Reuters reported in April 2024, and Apple has not confirmed the deal. Source: 9to5Mac, citing Reuters. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#deals-shutterstock-apple
$10 million was Google's winning bid in the August 2026 bankruptcy auction for Spirit Airlines' de-identified business data, with court approval pending, The Next Web reported in October 2026.
Source: The Next Web, August 2026On our site: Selling data from a closed companyLink
Cite this statistic
$10 million was Google's winning bid in the August 2026 bankruptcy auction for Spirit Airlines' de-identified business data, with court approval pending, The Next Web reported in October 2026. Source: The Next Web. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#deals-spirit-google
About $2.62 per minute on average, and up to about $7.25, is what Adobe paid for video to train its AI model, Bloomberg reported in April 2024.
Source: Bloomberg via BNN Bloomberg, April 2024On our site: Video training dataLink
Cite this statistic
About $2.62 per minute on average, and up to about $7.25, is what Adobe paid for video to train its AI model, Bloomberg reported in April 2024. Source: Bloomberg via BNN Bloomberg. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#deals-adobe-video
What do AI data buyers publish about payouts?
AI data buyers that publish payouts quote from $10,000 to more than $4 million per partnership, but only 11 of the 61 buyers we profile publish a dollar figure at all.
$100,000 to $500,000 for a qualified partnership, $500,000 to $3 million for large-scale or ongoing data and more than $3 million for highly unique data are the payment tiers on micro1's data partnerships page, checked November 2026.
Source: micro1, November 2026On our site: micro1Link
Cite this statistic
$100,000 to $500,000 for a qualified partnership, $500,000 to $3 million for large-scale or ongoing data and more than $3 million for highly unique data are the payment tiers on micro1's data partnerships page, checked November 2026. Source: micro1. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#buyers-micro1-tiers
$1bn over 12 months is what micro1 committed to buying and licensing operational data, financed by Citi and Hercules Capital Partners, The Next Web reported in October 2026.
Source: The Next Web, October 2026On our site: micro1Link
Cite this statistic
$1bn over 12 months is what micro1 committed to buying and licensing operational data, financed by Citi and Hercules Capital Partners, The Next Web reported in October 2026. Source: The Next Web. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#buyers-micro1-commitment
$10,000 to more than $1 million per partnership is the illustrative payout range on Scale AI's data partnership page, checked November 2026.
Source: Scale AI, November 2026On our site: Scale AILink
Cite this statistic
$10,000 to more than $1 million per partnership is the illustrative payout range on Scale AI's data partnership page, checked November 2026. Source: Scale AI. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#buyers-scale-range
$100,000 to more than $4 million is the headline payout range Handshake publishes for company data partnerships, checked November 2026.
Source: Handshake, November 2026On our site: Handshake AILink
Cite this statistic
$100,000 to more than $4 million is the headline payout range Handshake publishes for company data partnerships, checked November 2026. Source: Handshake. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#buyers-handshake-range
More than $20 million has been paid to licensors by Troveo, which says it works with more than 40 active buyers, according to its business data page checked November 2026.
Source: Troveo, November 2026On our site: TroveoLink
Cite this statistic
More than $20 million has been paid to licensors by Troveo, which says it works with more than 40 active buyers, according to its business data page checked November 2026. Source: Troveo. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#buyers-troveo-paid
11 of the 61 data buyers, marketplaces and brokers profiled in the Sell My Data To AI buyers directory publish a dollar figure for payouts or prices, as of November 2026.
Source: Data buyers directory, November 8, 2026On our site: Data buyers directoryLink
Cite this statistic
11 of the 61 data buyers, marketplaces and brokers profiled in the Sell My Data To AI buyers directory publish a dollar figure for payouts or prices, as of November 2026. Source: Sell My Data To AI. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#buyers-publish-dollar
How much are AI data contributors and experts paid?
AI data contributors with expert skills are offered up to $150 an hour or more on the platforms that publish rates, and TechCrunch reported an average above $85 an hour at Mercor. DataAnnotation lists generalist and language work at up to $50 and $40 an hour.
Over $85 an hour on average is what Mercor pays the experts on its platform, TechCrunch reported on October 27, 2025, when Mercor raised $350 million at a $10 billion valuation.
Source: TechCrunch, October 27, 2025On our site: MercorLink
Cite this statistic
Over $85 an hour on average is what Mercor pays the experts on its platform, TechCrunch reported on October 27, 2025, when Mercor raised $350 million at a $10 billion valuation. Source: TechCrunch. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#pay-mercor-hourly
More than $1.5 million a day is paid out by Mercor to a roster of more than 30,000 experts, TechCrunch reported on October 27, 2025.
Source: TechCrunch, October 27, 2025On our site: Sourcing expert and human-generated dataLink
Cite this statistic
More than $1.5 million a day is paid out by Mercor to a roster of more than 30,000 experts, TechCrunch reported on October 27, 2025. Source: TechCrunch. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#pay-mercor-daily
More than $500 million in payouts to more than 188,000 fellows is Handshake's own figure for its AI program, on its page checked November 2026.
Source: Handshake, November 2026On our site: Sourcing expert and human-generated dataLink
Cite this statistic
More than $500 million in payouts to more than 188,000 fellows is Handshake's own figure for its AI program, on its page checked November 2026. Source: Handshake. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#pay-handshake-payouts
Up to $150 or more an hour for coding work and up to $50 an hour for generalists are the rates DataAnnotation, the Surge AI affiliate, lists on its homepage, checked November 2026.
Source: DataAnnotation, November 2026On our site: Surge AILink
Cite this statistic
Up to $150 or more an hour for coding work and up to $50 an hour for generalists are the rates DataAnnotation, the Surge AI affiliate, lists on its homepage, checked November 2026. Source: DataAnnotation. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#pay-dataannotation-rates
Up to $125 or more an hour for legal, medical and finance experts, and up to $40 an hour for language specialists, are the rates DataAnnotation lists on its homepage, checked November 2026.
Source: DataAnnotation, November 2026On our site: Sourcing expert and human-generated dataLink
Cite this statistic
Up to $125 or more an hour for legal, medical and finance experts, and up to $40 an hour for language specialists, are the rates DataAnnotation lists on its homepage, checked November 2026. Source: DataAnnotation. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#pay-dataannotation-experts
$8.00 an hour is Prolific's minimum participant pay, with $12.00 an hour recommended, according to its pricing page checked November 2026.
Source: Prolific, November 2026On our site: ProlificLink
Cite this statistic
$8.00 an hour is Prolific's minimum participant pay, with $12.00 an hour recommended, according to its pricing page checked November 2026. Source: Prolific. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#pay-prolific-minimum
How much have AI training data lawsuits cost?
The largest disclosed AI training data settlement is $1.5 billion in Bartz v. Anthropic, and most of the cases we track are still pending.
$1.5 billion, about $3,000 per book, is what Anthropic agreed to pay in Bartz v. Anthropic, and the court gave the settlement final approval on July 20, 2026, according to the Association of American Publishers.
Source: Association of American Publishers, July 20, 2026On our site: AI training data lawsuits trackerLink
Cite this statistic
$1.5 billion, about $3,000 per book, is what Anthropic agreed to pay in Bartz v. Anthropic, and the court gave the settlement final approval on July 20, 2026, according to the Association of American Publishers. Source: Association of American Publishers. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#lawsuits-bartz-settlement
November 15, 2026 is the date by which the settlement administrator expected to send the first round of Bartz v. Anthropic payments, according to the official settlement website checked November 2026.
Source: JND Legal Administration, settlement administrator, November 2026On our site: AI training data lawsuits trackerLink
Cite this statistic
November 15, 2026 is the date by which the settlement administrator expected to send the first round of Bartz v. Anthropic payments, according to the official settlement website checked November 2026. Source: JND Legal Administration, settlement administrator. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#lawsuits-bartz-payments
23% of Clearview AI, which the judge estimated at $51.75 million, went to the class in the face-scan privacy settlement approved on March 20, 2025, Bloomberg Law reported.
Source: Bloomberg Law, March 20, 2025On our site: AI training data lawsuits trackerLink
Cite this statistic
23% of Clearview AI, which the judge estimated at $51.75 million, went to the class in the face-scan privacy settlement approved on March 20, 2025, Bloomberg Law reported. Source: Bloomberg Law. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#lawsuits-clearview-stake
More than $3 billion in damages is what music publishers say they could recover from Anthropic over more than 20,000 works allegedly downloaded by torrent, in a suit filed in January 2026, TechCrunch reported.
Source: TechCrunch, January 2026On our site: AI training data lawsuits trackerLink
Cite this statistic
More than $3 billion in damages is what music publishers say they could recover from Anthropic over more than 20,000 works allegedly downloaded by torrent, in a suit filed in January 2026, TechCrunch reported. Source: TechCrunch. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#lawsuits-music-publishers-anthropic
60,202 recordings are at issue in the further suit Universal and Sony filed against Suno in September 2026, alleging its v6 model built on earlier unlicensed models, Digital Music News reported.
Source: Digital Music News, September 2026On our site: AI training data lawsuits trackerLink
Cite this statistic
60,202 recordings are at issue in the further suit Universal and Sony filed against Suno in September 2026, alleging its v6 model built on earlier unlicensed models, Digital Music News reported. Source: Digital Music News. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#lawsuits-suno-recordings
Two of the three major record labels, Universal on October 29, 2025 and Warner in November 2025, settled their AI training claims against Udio, while Sony's case continues, according to Music Business Worldwide.
Source: Music Business Worldwide, November 2025On our site: AI training data lawsuits trackerLink
Cite this statistic
Two of the three major record labels, Universal on October 29, 2025 and Warner in November 2025, settled their AI training claims against Udio, while Sony's case continues, according to Music Business Worldwide. Source: Music Business Worldwide. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#lawsuits-udio-settlements
What have courts said about copyright and fair use in AI training?
US courts have reached different answers on fair use in AI training: two June 2025 rulings found training on books was fair use on their records, while the Third Circuit rejected fair use for Westlaw headnotes in September 2026.
Training Claude on lawfully bought books was fair use but keeping pirated copies from shadow libraries was not, Judge William Alsup held in Bartz v. Anthropic on June 23, 2025, according to Mondaq.
Source: Mondaq, June 23, 2025On our site: AI training data lawsuits trackerLink
Cite this statistic
Training Claude on lawfully bought books was fair use but keeping pirated copies from shadow libraries was not, Judge William Alsup held in Bartz v. Anthropic on June 23, 2025, according to Mondaq. Source: Mondaq. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#copyright-bartz-fair-use
Training Llama on 13 authors' books was fair use on the record because they had not shown market harm, Judge Vince Chhabria held in Kadrey v. Meta Platforms on June 25, 2025, while saying the ruling did not make training on copyrighted works lawful in general, according to Bar and Bench.
Source: Bar and Bench, June 25, 2025On our site: AI training data lawsuits trackerLink
Cite this statistic
Training Llama on 13 authors' books was fair use on the record because they had not shown market harm, Judge Vince Chhabria held in Kadrey v. Meta Platforms on June 25, 2025, while saying the ruling did not make training on copyrighted works lawful in general, according to Bar and Bench. Source: Bar and Bench. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#copyright-kadrey-fair-use
Using Westlaw headnotes to train a competing legal search tool was not fair use, the Third Circuit held on September 29, 2026 in Thomson Reuters v. Ross Intelligence (No. 25-2153), noting harm to a market for licensing content as AI training data, according to Jackson Walker.
Source: Jackson Walker, September 29, 2026On our site: AI training data lawsuits trackerLink
Cite this statistic
Using Westlaw headnotes to train a competing legal search tool was not fair use, the Third Circuit held on September 29, 2026 in Thomson Reuters v. Ross Intelligence (No. 25-2153), noting harm to a market for licensing content as AI training data, according to Jackson Walker. Source: Jackson Walker. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#copyright-ross-third-circuit
Summary judgment motions from both sides were filed on September 4, 2026 in the consolidated authors' and news publishers' cases against OpenAI and Microsoft before Judge Sidney Stein, according to the Authors Guild.
Source: Authors Guild, September 4, 2026On our site: AI training data lawsuits trackerLink
Cite this statistic
Summary judgment motions from both sides were filed on September 4, 2026 in the consolidated authors' and news publishers' cases against OpenAI and Microsoft before Judge Sidney Stein, according to the Authors Guild. Source: Authors Guild. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#copyright-openai-summary-judgment
Over 10,000 comments reached the US Copyright Office by December 2023 in response to its August 2023 notice of inquiry on copyright and artificial intelligence, according to the Office.
Source: US Copyright Office, December 2023On our site: Can you legally sell company data for AI training?Link
Cite this statistic
Over 10,000 comments reached the US Copyright Office by December 2023 in response to its August 2023 notice of inquiry on copyright and artificial intelligence, according to the Office. Source: US Copyright Office. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#copyright-office-comments
May 9, 2025 is when the US Copyright Office released a pre-publication version of Part 3 of its Copyright and Artificial Intelligence report, on generative AI training, with a final version still to come as of November 2026.
Source: US Copyright Office, May 9, 2025On our site: Can you legally sell company data for AI training?Link
Cite this statistic
May 9, 2025 is when the US Copyright Office released a pre-publication version of Part 3 of its Copyright and Artificial Intelligence report, on generative AI training, with a final version still to come as of November 2026. Source: US Copyright Office. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#copyright-office-part-3
GPT-4 and GPT-4o memorized the lyrics of nine German songs, and that memorization is a reproduction the text and data mining exceptions do not cover, the Munich Regional Court held on November 11, 2025 in GEMA v. OpenAI, according to Bird and Bird.
Outside the USSource: Bird and Bird, November 11, 2025On our site: AI training data lawsuits trackerLink
Cite this statistic
GPT-4 and GPT-4o memorized the lyrics of nine German songs, and that memorization is a reproduction the text and data mining exceptions do not cover, the Munich Regional Court held on November 11, 2025 in GEMA v. OpenAI, according to Bird and Bird. Source: Bird and Bird. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#copyright-gema-openai
Which US privacy laws apply when you sell data for AI training?
US privacy law for AI training data is mostly state law: 19 states have comprehensive privacy laws in force, and California's regulator, the California Privacy Protection Agency, sets CCPA fines and runs a deletion platform for data brokers.
19 US states have comprehensive consumer privacy laws in force as of November 2026 and 3 more take effect in 2027, by our count from the Future of Privacy Forum's effective dates chart and later state enactments.
Source: Future of Privacy Forum, November 8, 2026On our site: Data privacy laws for AI training by countryLink
Cite this statistic
19 US states have comprehensive consumer privacy laws in force as of November 2026 and 3 more take effect in 2027, by our count from the Future of Privacy Forum's effective dates chart and later state enactments. Source: Future of Privacy Forum. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#privacy-state-laws
$2,663 per violation and $7,988 per intentional violation, or violation involving consumers under 16, are the CCPA's maximum administrative fines from January 1, 2025, according to the California Privacy Protection Agency.
Source: California Privacy Protection Agency, January 1, 2025On our site: Selling company data to AI in CaliforniaLink
Cite this statistic
$2,663 per violation and $7,988 per intentional violation, or violation involving consumers under 16, are the CCPA's maximum administrative fines from January 1, 2025, according to the California Privacy Protection Agency. Source: California Privacy Protection Agency. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#privacy-ccpa-fines
$26,625,000 in annual gross revenue is the CCPA's revenue threshold for a business from January 1, 2025, up from $25,000,000, according to the California Privacy Protection Agency.
Source: California Privacy Protection Agency, January 1, 2025On our site: Selling company data to AI in CaliforniaLink
Cite this statistic
$26,625,000 in annual gross revenue is the CCPA's revenue threshold for a business from January 1, 2025, up from $25,000,000, according to the California Privacy Protection Agency. Source: California Privacy Protection Agency. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#privacy-ccpa-threshold
$1.35 million is the fine the California Privacy Protection Agency announced against Tractor Supply Company on September 30, 2025, the largest in the agency's history, in a case that included failing to offer a working way to opt out of data sales and sharing.
Source: California Privacy Protection Agency, September 30, 2025On our site: Selling company data to AI in CaliforniaLink
Cite this statistic
$1.35 million is the fine the California Privacy Protection Agency announced against Tractor Supply Company on September 30, 2025, the largest in the agency's history, in a case that included failing to offer a working way to opt out of data sales and sharing. Source: California Privacy Protection Agency. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#privacy-tractor-supply
300,000 Californians had signed up for the state's Delete Request and Opt-out Platform (DROP) by June 2026, and registered data brokers reached a record high, according to CalPrivacy.
Source: CalPrivacy (California Privacy Protection Agency), June 2026On our site: Selling company data to AI in CaliforniaLink
Cite this statistic
300,000 Californians had signed up for the state's Delete Request and Opt-out Platform (DROP) by June 2026, and registered data brokers reached a record high, according to CalPrivacy. Source: CalPrivacy (California Privacy Protection Agency). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#privacy-drop-signups
January 1, 2026 is when the California Privacy Protection Agency's regulations took effect requiring a risk assessment before a business sells or shares personal information, according to the agency.
Source: California Privacy Protection Agency, January 1, 2026On our site: Selling company data to AI in CaliforniaLink
Cite this statistic
January 1, 2026 is when the California Privacy Protection Agency's regulations took effect requiring a risk assessment before a business sells or shares personal information, according to the agency. Source: California Privacy Protection Agency. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#privacy-cppa-risk-assessments
January 1, 2026 is when California's AB 2013 took effect, requiring developers of generative AI systems offered to the public to post a summary of their training data, including whether datasets were purchased or licensed, according to Goodwin Procter.
Source: Goodwin Procter, January 1, 2026On our site: Selling company data to AI in CaliforniaLink
Cite this statistic
January 1, 2026 is when California's AB 2013 took effect, requiring developers of generative AI systems offered to the public to post a summary of their training data, including whether datasets were purchased or licensed, according to Goodwin Procter. Source: Goodwin Procter. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#privacy-ab-2013
January 1, 2026 is when Texas House Bill 149 took effect, adding an AI training exemption to the state's biometric identifier law, CUBI, unless a system identifies specific people, according to the Texas Business and Commerce Code.
Source: Texas Legislature, January 1, 2026On our site: Selling company data to AI in TexasLink
Cite this statistic
January 1, 2026 is when Texas House Bill 149 took effect, adding an AI training exemption to the state's biometric identifier law, CUBI, unless a system identifies specific people, according to the Texas Business and Commerce Code. Source: Texas Legislature. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#privacy-texas-cubi
January 1, 2027 is when Colorado's SB26-189 takes effect, after it repealed and replaced the 2024 Colorado AI Act in May 2026, according to Lathrop GPM.
Source: Lathrop GPM, May 2026On our site: Selling company data to AI in ColoradoLink
Cite this statistic
January 1, 2027 is when Colorado's SB26-189 takes effect, after it repealed and replaced the 2024 Colorado AI Act in May 2026, according to Lathrop GPM. Source: Lathrop GPM. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#privacy-colorado-sb189
Is AI running out of training data?
AI developers are on course to use up the stock of public human-written text between 2026 and 2032, according to Epoch AI, which put that stock at about 300 trillion tokens in June 2024.
About 300 trillion tokens is Epoch AI's estimate of the effective stock of human-generated public text, with a 90% confidence interval of 100 trillion to 1,000 trillion, in its paper published June 6, 2024.
EstimateSource: Epoch AI (Villalobos et al.), June 6, 2024On our site: What data do AI companies buy?Link
Cite this statistic
About 300 trillion tokens is Epoch AI's estimate of the effective stock of human-generated public text, with a 90% confidence interval of 100 trillion to 1,000 trillion, in its paper published June 6, 2024. Source: Epoch AI (Villalobos et al.). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#scarcity-epoch-stock
2026 to 2032 is the window in which language models trained on current trends will have used the full stock of public human text, according to Epoch AI's estimate of June 6, 2024 (80% confidence interval).
EstimateSource: Epoch AI (Villalobos et al.), June 6, 2024On our site: What data do AI companies buy?Link
Cite this statistic
2026 to 2032 is the window in which language models trained on current trends will have used the full stock of public human text, according to Epoch AI's estimate of June 6, 2024 (80% confidence interval). Source: Epoch AI (Villalobos et al.). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#scarcity-epoch-window
2027 is when that stock would be fully used if models are overtrained by a factor of 5, Epoch AI estimated on June 6, 2024, noting that Llama 3 70B was overtrained by a factor of 10.
EstimateSource: Epoch AI (Villalobos et al.), June 6, 2024On our site: What data do AI companies buy?Link
Cite this statistic
2027 is when that stock would be fully used if models are overtrained by a factor of 5, Epoch AI estimated on June 6, 2024, noting that Llama 3 70B was overtrained by a factor of 10. Source: Epoch AI (Villalobos et al.). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#scarcity-epoch-overtraining
Over 15 trillion tokens from publicly available sources were used to pretrain Llama 3, a dataset seven times larger than Llama 2's with four times more code, Meta said on April 18, 2024.
Source: Meta AI, April 18, 2024On our site: What data do AI companies buy?Link
Cite this statistic
Over 15 trillion tokens from publicly available sources were used to pretrain Llama 3, a dataset seven times larger than Llama 2's with four times more code, Meta said on April 18, 2024. Source: Meta AI. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#scarcity-llama-3-tokens
Over 300 billion pages spanning 15 years are in the Common Crawl corpus, with 3 to 5 billion new pages added each month, according to Common Crawl's homepage checked November 2026.
Source: Common Crawl Foundation, November 2026On our site: Data provenance explainedLink
Cite this statistic
Over 300 billion pages spanning 15 years are in the Common Crawl corpus, with 3 to 5 billion new pages added each month, according to Common Crawl's homepage checked November 2026. Source: Common Crawl Foundation. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#scarcity-common-crawl
How much AI training data is synthetic?
Synthetic data made up 40% of the pre-training mix for Microsoft's Phi-4 and over 98% of the alignment data for Nvidia's Nemotron-4 340B, but research on model collapse shows that keeping real data in the mix still matters.
40% of the pre-training mix for Microsoft's Phi-4 was synthetic data, alongside 15% web, 15% web rewrites, 20% code and 10% acquired sources, according to the Phi-4 technical report of December 2024.
Source: arXiv (Abdin et al., Microsoft, 2024), December 2024On our site: Synthetic data vs licensed real dataLink
Cite this statistic
40% of the pre-training mix for Microsoft's Phi-4 was synthetic data, alongside 15% web, 15% web rewrites, 20% code and 10% acquired sources, according to the Phi-4 technical report of December 2024. Source: arXiv (Abdin et al., Microsoft, 2024). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#synthetic-phi-4
Over 98% of the data used to align Nvidia's Nemotron-4 340B was synthetically generated, according to Nvidia's technical report of June 2024.
Source: arXiv (Nvidia, 2024), June 2024On our site: Synthetic data vs licensed real dataLink
Cite this statistic
Over 98% of the data used to align Nvidia's Nemotron-4 340B was synthetically generated, according to Nvidia's technical report of June 2024. Source: arXiv (Nvidia, 2024). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#synthetic-nemotron
Keeping a random 10% of the original data each generation led to only minor degradation when a language model was fine-tuned over generations, Shumailov and colleagues reported in Nature in July 2024.
Source: Nature (Shumailov et al., July 2024), July 2024On our site: Synthetic data vs licensed real dataLink
Cite this statistic
Keeping a random 10% of the original data each generation led to only minor degradation when a language model was fine-tuned over generations, Shumailov and colleagues reported in Nature in July 2024. Source: Nature (Shumailov et al., July 2024). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#synthetic-shumailov
Even 1% synthetic data can lead to model collapse in a supervised regression setting, where larger training sets stop improving performance, Dohmatob and colleagues showed in their October 2024 paper on strong model collapse.
Source: arXiv (Dohmatob et al., 2024), October 2024On our site: Synthetic data vs licensed real dataLink
Cite this statistic
Even 1% synthetic data can lead to model collapse in a supervised regression setting, where larger training sets stop improving performance, Dohmatob and colleagues showed in their October 2024 paper on strong model collapse. Source: arXiv (Dohmatob et al., 2024). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#synthetic-strong-collapse
2030 is the year by which synthetic data will completely overshadow real data in AI models, Gartner predicted on June 22, 2022.
EstimateSource: Gartner, June 22, 2022On our site: Synthetic data vs licensed real dataLink
Cite this statistic
2030 is the year by which synthetic data will completely overshadow real data in AI models, Gartner predicted on June 22, 2022. Source: Gartner. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#synthetic-gartner-2030
How many websites and datasets restrict AI training?
A growing share of the web and of public datasets restricts AI training: 45% of C4 is restricted by terms of service, and more than 1 million Cloudflare customers block AI crawlers.
Above 70% license omission rates and above 50% error rates were found on popular dataset hosting sites when the Data Provenance Initiative audited more than 1,800 text datasets, as reported in Nature Machine Intelligence in 2024.
Source: MIT Media Lab (Nature Machine Intelligence, 2024), 2024On our site: Data provenance explainedLink
Cite this statistic
Above 70% license omission rates and above 50% error rates were found on popular dataset hosting sites when the Data Provenance Initiative audited more than 1,800 text datasets, as reported in Nature Machine Intelligence in 2024. Source: MIT Media Lab (Nature Machine Intelligence, 2024). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#provenance-dpi-audit
5% or more of all tokens in C4, and 28% or more of its most actively maintained critical sources, became fully restricted from use in a single year from 2023 to 2024, the Data Provenance Initiative reported in July 2024.
Source: arXiv (Longpre et al., Data Provenance Initiative, 2024), July 20, 2024On our site: Data provenance explainedLink
Cite this statistic
5% or more of all tokens in C4, and 28% or more of its most actively maintained critical sources, became fully restricted from use in a single year from 2023 to 2024, the Data Provenance Initiative reported in July 2024. Source: arXiv (Longpre et al., Data Provenance Initiative, 2024). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#provenance-consent-in-crisis-tokens
45% of C4 is restricted by terms of service crawling restrictions, the Data Provenance Initiative reported in July 2024.
Source: arXiv (Longpre et al., Data Provenance Initiative, 2024), July 20, 2024On our site: Data provenance explainedLink
Cite this statistic
45% of C4 is restricted by terms of service crawling restrictions, the Data Provenance Initiative reported in July 2024. Source: arXiv (Longpre et al., Data Provenance Initiative, 2024). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#provenance-consent-in-crisis-terms
Over 80% of the source content in widely used text, speech and video datasets carries non-commercial restrictions, while less than 33% of the datasets are restrictively licensed, the Data Provenance Initiative found in December 2024 after reviewing nearly 4,000 public datasets.
Source: arXiv (Longpre et al., Data Provenance Initiative, 2024), December 19, 2024On our site: Data provenance explainedLink
Cite this statistic
Over 80% of the source content in widely used text, speech and video datasets carries non-commercial restrictions, while less than 33% of the datasets are restrictively licensed, the Data Provenance Initiative found in December 2024 after reviewing nearly 4,000 public datasets. Source: arXiv (Longpre et al., Data Provenance Initiative, 2024). Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#provenance-multimodal-restrictions
More than 1 million Cloudflare customers had chosen to block AI crawlers since a one-click option launched in September 2024, Cloudflare said on July 1, 2025, when it began blocking AI crawlers by default.
Source: Cloudflare, July 1, 2025On our site: Data provenance explainedLink
Cite this statistic
More than 1 million Cloudflare customers had chosen to block AI crawlers since a one-click option launched in September 2024, Cloudflare said on July 1, 2025, when it began blocking AI crawlers by default. Source: Cloudflare. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#provenance-cloudflare-blocking
38,000 pages crawled for every referred page visit was Anthropic's crawl-to-refer ratio in July 2025, against about 1,091 for OpenAI, Cloudflare reported on August 29, 2025.
Source: Cloudflare, August 29, 2025On our site: Data provenance explainedLink
Cite this statistic
38,000 pages crawled for every referred page visit was Anthropic's crawl-to-refer ratio in July 2025, against about 1,091 for OpenAI, Cloudflare reported on August 29, 2025. Source: Cloudflare. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#provenance-crawl-to-refer
Robots.txt is a request, not a technological measure, so ignoring it is not circumvention under the DMCA, the court overseeing the OpenAI copyright cases held on December 15, 2025 in Ziff Davis v. OpenAI, according to its opinion and order.
Source: U.S. District Court for the Southern District of New York, via OpenAI, December 15, 2025On our site: AI training data lawsuits trackerLink
Cite this statistic
Robots.txt is a request, not a technological measure, so ignoring it is not circumvention under the DMCA, the court overseeing the OpenAI copyright cases held on December 15, 2025 in Ziff Davis v. OpenAI, according to its opinion and order. Source: U.S. District Court for the Southern District of New York, via OpenAI. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#provenance-robots-txt-ruling
What do businesses and the public think about AI training data?
Surveys of businesses and the public about AI training data show caution on both sides: 27% of organizations in Cisco's study had banned generative AI at least temporarily, and 87% of US adults in a TELUS Digital survey want companies to be open about data sourcing.
27% of organizations had banned generative AI at least temporarily over privacy and data security risks, according to Cisco's 2024 Data Privacy Benchmark Study of 2,600 privacy and security professionals, published January 25, 2024.
Source: Cisco, January 25, 2024On our site: Is selling company data to AI safe?Link
Cite this statistic
27% of organizations had banned generative AI at least temporarily over privacy and data security risks, according to Cisco's 2024 Data Privacy Benchmark Study of 2,600 privacy and security professionals, published January 25, 2024. Source: Cisco. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#attitudes-cisco-bans
48% of privacy and security professionals admitted entering non-public company information into generative AI tools, and 63% said their organizations limit what data can be entered, Cisco reported on January 25, 2024.
Source: Cisco, January 25, 2024On our site: Is selling company data to AI safe?Link
Cite this statistic
48% of privacy and security professionals admitted entering non-public company information into generative AI tools, and 63% said their organizations limit what data can be entered, Cisco reported on January 25, 2024. Source: Cisco. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#attitudes-cisco-non-public
87% of US adults familiar with generative AI say companies should be transparent about how they source data for GenAI models, up from 75% in 2023, according to a TELUS Digital survey of 1,000 adults published June 24, 2025.
Source: TELUS Digital via Business Wire, June 24, 2025On our site: Telling staff and clients about a data licenseLink
Cite this statistic
87% of US adults familiar with generative AI say companies should be transparent about how they source data for GenAI models, up from 75% in 2023, according to a TELUS Digital survey of 1,000 adults published June 24, 2025. Source: TELUS Digital via Business Wire. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#attitudes-telus-transparency
81% of Americans familiar with AI say companies' use of it will lead to their personal information being used in ways they are not comfortable with, Pew Research Center reported on October 18, 2023.
Source: Pew Research Center, October 18, 2023On our site: Telling staff and clients about a data licenseLink
Cite this statistic
81% of Americans familiar with AI say companies' use of it will lead to their personal information being used in ways they are not comfortable with, Pew Research Center reported on October 18, 2023. Source: Pew Research Center. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#attitudes-pew-discomfort
20% of news leaders expect substantial future revenue from AI licensing deals, 49% expect a minor contribution and 20% expect none, the Reuters Institute found in a survey of 280 digital leaders in 51 countries published January 12, 2026.
Source: Reuters Institute for the Study of Journalism, January 12, 2026On our site: AI data deals trackerLink
Cite this statistic
20% of news leaders expect substantial future revenue from AI licensing deals, 49% expect a minor contribution and 20% expect none, the Reuters Institute found in a survey of 280 digital leaders in 51 countries published January 12, 2026. Source: Reuters Institute for the Study of Journalism. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#attitudes-reuters-institute-licensing
Which platform terms restrict AI training on your data?
Platform terms that restrict AI training mostly target data pulled through APIs: 5 of the 12 SaaS platforms we reviewed restrict it, and X bars training foundation models on its content.
5 of the 12 SaaS platforms in the Sell My Data To AI platform data rights review restrict AI training on data pulled through their APIs (Slack, Google Workspace, Zendesk, HubSpot and Zoom), as of November 2026.
Source: Platform data rights, November 8, 2026On our site: Platform data rightsLink
Cite this statistic
5 of the 12 SaaS platforms in the Sell My Data To AI platform data rights review restrict AI training on data pulled through their APIs (Slack, Google Workspace, Zendesk, HubSpot and Zoom), as of November 2026. Source: Sell My Data To AI. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#platforms-api-restrictions
Slack's API Terms of Service bar apps offered outside your organization from using API Data to train a large language model, in the version effective October 10, 2025, according to Slack.
Source: Slack, October 10, 2025On our site: Slack data rightsLink
Cite this statistic
Slack's API Terms of Service bar apps offered outside your organization from using API Data to train a large language model, in the version effective October 10, 2025, according to Slack. Source: Slack. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#platforms-slack-api
Google's Workspace API User Data and Developer Policy prohibits using user data to create, train or improve a machine learning or artificial intelligence model beyond that user's own personalized model, in the version updated September 3, 2026.
Source: Google for Developers, September 3, 2026On our site: Google Workspace data rightsLink
Cite this statistic
Google's Workspace API User Data and Developer Policy prohibits using user data to create, train or improve a machine learning or artificial intelligence model beyond that user's own personalized model, in the version updated September 3, 2026. Source: Google for Developers. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#platforms-google-workspace-api
Zendesk's Developer Terms bar using API Data for creating, training or improving any general AI model, under section 3.6(iii) of the version effective August 12, 2025.
Source: Zendesk, August 12, 2025On our site: Zendesk data rightsLink
Cite this statistic
Zendesk's Developer Terms bar using API Data for creating, training or improving any general AI model, under section 3.6(iii) of the version effective August 12, 2025. Source: Zendesk. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#platforms-zendesk-api
HubSpot's Developer Terms say developers must not use Customer Data or Content to train, fine-tune or improve AI, under section 4.B.8 of the version last modified May 4, 2026.
Source: HubSpot, May 4, 2026On our site: HubSpot data rightsLink
Cite this statistic
HubSpot's Developer Terms say developers must not use Customer Data or Content to train, fine-tune or improve AI, under section 4.B.8 of the version last modified May 4, 2026. Source: HubSpot. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#platforms-hubspot-api
Zoom's API License and Terms of Use bar using Customer Content to train, develop, evaluate, test, improve or modify AI or machine learning models without Zoom's express written permission, under section 2.8 of the version last updated July 16, 2025.
Source: Zoom, July 16, 2025On our site: Zoom data rightsLink
Cite this statistic
Zoom's API License and Terms of Use bar using Customer Content to train, develop, evaluate, test, improve or modify AI or machine learning models without Zoom's express written permission, under section 2.8 of the version last updated July 16, 2025. Source: Zoom. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#platforms-zoom-api
X's developer agreement bars using the X API or X Content to fine-tune or train a foundation or frontier model, a clause added in June 2025, TechCrunch reported on June 5, 2025.
Source: TechCrunch, June 5, 2025On our site: Platform data rightsLink
Cite this statistic
X's developer agreement bars using the X API or X Content to fine-tune or train a foundation or frontier model, a clause added in June 2025, TechCrunch reported on June 5, 2025. Source: TechCrunch. Via Sell My Data To AI, https://sellmydatatoai.net/ai-training-data-statistics#platforms-x-developer-agreement
How to cite these statistics
You can reuse this page and its data under CC BY 4.0. Please credit Sell My Data To AI, link to this page and cite the original source named in each statistic. Every statistic has its own link and a ready-made citation. A suggested citation for the page:
Sell My Data To AI (2026). AI training data statistics 2027. https://sellmydatatoai.net/ai-training-data-statistics
- URL
- https://sellmydatatoai.net/ai-training-data-statistics
- Data
- Download the statistics (CSV)
- Last checked
- November 2026
- License
- Creative Commons Attribution 4.0 (CC BY 4.0)
How were these statistics chosen?
Sell My Data To AI compiled these 85 statistics from primary sources first: company filings and announcements, court records, regulators, research papers and named survey publishers. Figures we already cover are linked to the page on this site that explains them. Every statistic was last checked in November 2026.
- A statistic is included only if a named publisher states it in a document we could read, and the sentence names that publisher and gives a date.
- Each figure is quoted as published. We do not round in a way that changes meaning, and we do not average, combine or extrapolate.
- Estimates, forecasts and predictions are labeled, with the firm and the year named in the sentence.
- Reported deal values come from the outlet that reported them, and we say when a party has not confirmed a deal.
- Where only a secondary report exists, such as a news report of a private deal, the sentence names who reported it.
- Figures already on this site are checked against the page that covers them each time the site is built, and a figure that no longer matches is removed.
- Counts from our own datasets, such as deals with a disclosed value, are computed from those datasets when the site is built.
- US sources come first. Figures from outside the US are labeled.
We update this page when a source changes or a better primary source appears. We are a broker, not a law firm. We tell you what buyers will check and help you prepare, but you should take independent legal advice before signing any license.
Read more about how we research, or see the datasets behind many of these figures: the AI data price index and the State of AI data licensing 2027 report.
Frequently asked questions
How big is the AI training data market?
What is the largest AI data licensing deal?
Will AI run out of training data?
How much do AI companies pay data contributors and experts?
Can I cite or reuse these statistics?
Sources
- AI training dataset market size and share report, 2026 to 2033, Grand View Research, accessed November 2026
- AI training dataset market size, share and 2031 growth trends report, Mordor Intelligence, accessed November 2026
- The rise of AI training dataset market: a $9.58 billion industry (MarketsandMarkets), MarketsandMarkets via GlobeNewswire, accessed November 2026
- Reddit, Inc. Form S-1 registration statement, US Securities and Exchange Commission (EDGAR), accessed November 2026
- Reddit, Inc. Form 10-K for fiscal year 2024, US Securities and Exchange Commission (EDGAR), accessed November 2026
- Shutterstock, Inc. Form 10-K for fiscal year 2023, US Securities and Exchange Commission (EDGAR), accessed November 2026
- Informa PLC 2024 preliminary full-year results, Informa PLC via the FCA National Storage Mechanism, accessed November 2026
- Exclusive: Scale AI's bigger rival Surge AI seeks up to $1 billion capital raise, sources say, Reuters via The Star, accessed November 2026
- Reported deal: News Corp and OpenAI, Engadget, citing The Wall Street Journal, accessed November 2026
- Reported deal: Reddit and Google, Reuters via Nieman Lab, accessed November 2026
- Reported deal: The New York Times and Amazon, Engadget, citing The Wall Street Journal, accessed November 2026
- Reported deal: Dotdash Meredith and OpenAI, Engadget, citing Adweek, accessed November 2026
- Reported deal: HarperCollins and Microsoft, Publishers Weekly, accessed November 2026
- Reported deal: Shutterstock and Apple, 9to5Mac, citing Reuters, accessed November 2026
- Reported deal: Spirit Airlines and Google, The Next Web, accessed November 2026
- Reported deal: A shut-down crypto startup and an unnamed buyer, MIT Technology Review, accessed November 2026
- Adobe is buying videos for $3 per minute to build AI model, Bloomberg via BNN Bloomberg, accessed November 2026
- Data partnerships, micro1, accessed November 2026
- micro1 commits $1bn to enterprise data for AI training, The Next Web, accessed November 2026
- Data partnership, Scale AI, accessed November 2026
- Data partnerships, Handshake, accessed November 2026
- Track record, Polyshares, accessed November 2026
- Business data, Troveo, accessed November 2026
- Perplexity says it will give publishers 80% cut of new subscription product's revenue, Axios, accessed November 2026
- Mercor quintuples valuation to $10B with $350M Series C, TechCrunch, accessed November 2026
- Handshake AI program, Handshake, accessed November 2026
- DataAnnotation homepage, DataAnnotation, accessed November 2026
- Pricing, Prolific, accessed November 2026
- AAP welcomes court's final settlement approval in Bartz v. Anthropic, Association of American Publishers, accessed November 2026
- Anthropic Copyright Settlement website (Bartz v. Anthropic PBC, 3:24-cv-5417), JND Legal Administration, settlement administrator, accessed November 2026
- Clearview AI Gets Settlement Approved in Face-Scan Privacy Case, Bloomberg Law, accessed November 2026
- Music publishers sue Anthropic for $3B over 'flagrant piracy' of 20,000 works, TechCrunch, accessed November 2026
- Suno blasts latest UMG/Sony lawsuit as 'fundamentally flawed on both the facts and the law', Digital Music News, accessed November 2026
- Universal Music settles Udio lawsuit, strikes deal for licensed AI music platform, Music Business Worldwide, accessed November 2026
- District court issues AI fair use decision, Mondaq, accessed November 2026
- AI training may be fair use but using pirated books is not: US courts in Meta, Claude cases, Bar and Bench, accessed November 2026
- Training Day: Third Circuit affirms Westlaw headnotes decision in AI copyright fair use case, Jackson Walker, accessed November 2026
- Plaintiffs file motion for summary judgment v. OpenAI and Microsoft, Authors Guild, accessed November 2026
- Copyright and artificial intelligence, US Copyright Office, accessed November 2026
- Landmark ruling of the Munich Regional Court (GEMA v OpenAI) on copyright and AI training, Bird and Bird, accessed November 2026
- Effective dates chart, US state comprehensive privacy laws, Future of Privacy Forum, accessed November 2026
- Updated monetary thresholds in CCPA, California Privacy Protection Agency, accessed November 2026
- Nation's largest rural lifestyle retailer to pay $1.35M over CCPA violations, California Privacy Protection Agency, accessed November 2026
- Privacy momentum builds: 300,000 Californians sign up for DROP as registered data brokers hit a record high, CalPrivacy (California Privacy Protection Agency), accessed November 2026
- CCPA updates, cybersecurity audits, risk assessments, automated decisionmaking technology and insurance companies, California Privacy Protection Agency, accessed November 2026
- California's AB 2013 takes effect, Goodwin Procter, accessed November 2026
- Texas Business and Commerce Code chapter 503: biometric identifiers, Texas Legislature, accessed November 2026
- Colorado enacts new law regulating automated decision-making technology, Lathrop GPM, accessed November 2026
- Will we run out of data? Limits of LLM scaling based on human-generated data, Epoch AI (Villalobos et al.), accessed November 2026
- Introducing Meta Llama 3: the most capable openly available LLM to date, Meta AI, accessed November 2026
- Common Crawl homepage, Common Crawl Foundation, accessed November 2026
- Phi-4 technical report, arXiv (Abdin et al., Microsoft, 2024), accessed November 2026
- Nemotron-4 340B technical report, arXiv (Nvidia, 2024), accessed November 2026
- AI models collapse when trained on recursively generated data, Nature (Shumailov et al., July 2024), accessed November 2026
- Strong model collapse, arXiv (Dohmatob et al., 2024), accessed November 2026
- Is synthetic data the future of AI?, Gartner, accessed November 2026
- A large-scale audit of dataset licensing and attribution in AI, MIT Media Lab (Nature Machine Intelligence, 2024), accessed November 2026
- Consent in crisis: the rapid decline of the AI data commons, arXiv (Longpre et al., Data Provenance Initiative, 2024), accessed November 2026
- Bridging the data provenance gap across text, speech and video, arXiv (Longpre et al., Data Provenance Initiative, 2024), accessed November 2026
- Cloudflare just changed how AI crawlers scrape the internet at large, Cloudflare, accessed November 2026
- The crawl-to-click gap: Cloudflare data on AI bots, training, and referrals, Cloudflare, accessed November 2026
- Opinion and order, December 15, 2025 (Dkt. 968), U.S. District Court for the Southern District of New York, via OpenAI, accessed November 2026
- More than 1 in 4 organizations banned use of GenAI over privacy and data security risks: new Cisco study, Cisco, accessed November 2026
- Not all datasets are created equal: new TELUS Digital survey shows trust in AI is dependent on how data is sourced, TELUS Digital via Business Wire, accessed November 2026
- How Americans view data privacy, Pew Research Center, accessed November 2026
- AI survey: 90 percent of writers believe authors should be compensated for AI training use, Authors Guild, accessed November 2026
- Journalism, media, and technology trends and predictions 2026, Reuters Institute for the Study of Journalism, accessed November 2026
- Slack API Terms of Service, Slack, accessed November 2026
- Google Workspace API User Data and Developer Policy, Google for Developers, accessed November 2026
- Zendesk Developer Terms, Zendesk, accessed November 2026
- HubSpot Developer Terms, HubSpot, accessed November 2026
- Zoom API License and Terms of Use, Zoom, accessed November 2026
- X changes its terms to bar training of AI models using its content, TechCrunch, accessed November 2026
For data owners
Find out what your data could be worth
Published figures are ceilings, not quotes. A free valuation gives you a view of likely buyers, a realistic range and the rights issues to clear before a sale. Sellers pay nothing to be assessed.
For AI teams
Tell us what training data you need
Submit a written requirement and we find, vet and shortlist sellers of rights-cleared training data. Buyers pay nothing to submit a requirement or review a shortlist.