DataPacts

AI Training Data Statistics

Checked

At a glance

AI training data has no official statistics: the figures on the record are counts from public trackers, amounts stated by parties and filings, estimates published by research firms and totals that buyers claim, and this page lists each with who states it and as of when.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

The deal tracker holds 74 deals and the case tracker 70 rows as of 2026-11-09, the program terms table holds 143 programs, five publishers state eight market estimates, and eight companies state a running paid-out total.

Counts are the easy part and amounts are thin: nine of the 74 deals carry an amount stated by a party or a filing, and no seller has confirmed an amount received under any program. These numbers are one measurable part of AI training data licensing, and each keeps its source and date.

Fact-check note: every figure was taken from a live page of this site, which cites its primary source, and is dated as described in how we source, date and check every figure.

What Counts as a Statistic on This Page, and How Each One Is Labeled

A statistic here is a number that already sits on a live page of this site with its source, labeled by who states it. No figure is averaged, blended or summed across sources.

Five labels appear in the tables:

  • Tracker count: rows counted in the deal, case or program tables, not deals or lawsuits in the world.
  • Stated amount: an amount a party, a filing or a court record states, with that source named.
  • Publisher's estimate: a research firm's model of a market it defines itself.
  • Buyer's claim: a total a company publishes about its own payments, unconfirmed by any recipient.
  • Computed difference: one stated figure subtracted from another, never an average.

The rules for sourcing and dating follow how we source, date and check every figure; where one source states two figures, both stay.

Key Figures on the Record, Dated

The table indexes the counts and headline figures in the market section, each with the date it holds and the type of source behind it.

Statistic Value As of Source type
Deals in the deal tracker 74 rows 2026-11-09, counted 2026-11-17 site tracker
Tracker deals with an amount stated by a party or filing 9 2026-11-09 filings, dockets
Tracker deals marked not disclosed (n/d) by any party or filing 55 2026-11-09 site tracker
Rows in the case tracker 70, in 12 jurisdictions rows 2026-11-09, statuses to 2026-11-16, counted 2026-11-17 court sources
Training rulings on record 9 2026-11-09 court sources
Programs in the terms table 143 (140 recorded, 3 not verified) 2026-11-09 operators' pages
Contradicting value pairs on operators' own pages 13 2026-11-09 operators' pages
Companies in the company data program directory 41 2026-11-09 operators' pages
Market-size estimates of the AI training dataset market 8, from 5 publishers reports 2025-05 to 2026-11-02 publishers' pages, claims
AI-related revenue figures filed by companies 28, from 8 filers 2026-11-09 securities filings
Per-unit prices on record 22 rows 2026-11-09 filings, courts, vendor pages
Companies in the funding table; stated valuations 22; 6 2026-11-09 (Snorkel AI 2026-11-16) companies' statements
Documented changes to buyer program terms 4 2026-11-10 operators' pages
Companies publishing a paid-out total 8 2026-11-09 (Handshake AI, Troveo 2026-11-10) payers' claims

A tracker count measures what this site recorded, a filing what a company reported, and a claim what a company says about itself.

Deal Counts by Year, Data Type and Stated Use

The deal tracker holds 74 deals announced from 2022-10-25 to 2026-10-22, as of 2026-11-09, and every count below counts its rows, not deals in the world. Four program announcements naming 21 publisher rows are kept apart from the 74.

Announcements per Year and per Data Type Announcements rose from 1 in 2022 to 27 in 2024, then 24 in 2025 and 20 in 2026 up to November 9, and news publishers account for 43 of the 74 deals. Program announcements name several publishers on one day, among them the Publisher Content Marketplace (7 rows, 2026-02-03) and Copilot Daily (5 rows, 2024-10-01).

Tracker rows as of 2026-11-09, counted 2026-11-17.

Announced Deals
2022 1
2023 2
2024 27
2025 24
2026 (to November 9) 20
Total 74

{#deal-types}

Data type Deals
News 43
Platform 8
Music 7
Books 4
Clinical 4
Images 3
Video 3
Code 1
Company 1
Total 74

What Each Deal Is Stated to Cover

Sixteen of the 74 deals state training or model improvement, 32 cover display, grounding or answers without stating training, 1 excludes training and 25 do not state the use. Not stated does not mean the contract forbids training.

As of 2026-11-09, from the parties' own words.

Stated use Deals
T: training or model improvement stated 16
D: display, grounding or answers, training not stated 32
X: training excluded (Le Monde and Perplexity) 1
n/s: use not stated 25
Total 74

Amounts Stated by a Party or a Filing

Nine of the 74 deals carry an amount stated by a party or a filing, and 55 are marked not disclosed, with no amount from any party or filing. HarperCollins and Disney have no amount stated by any party.

The nine stated amounts, newest statement first, as of 2026-11-09, with no total row.

Deal Amount as stated Stated by Date What it is not Source
Spirit Airlines business data USD 10,000,000 successful bid (Google LLC); USD 7,500,000 alternate bid (Mercor.io Corporation) Bankr. S.D.N.Y. 25-11897 (SHL), Dkt. 1463 2026-08-14 a bid, not a sale; not approved as of 2026-11-09, hearing set for 2026-11-17 (Dkt. 1779) courtlistener.com
Tempus AI and Pathos AI, neurology up to USD 55.0 million: USD 100 per delivered record (USD 50.0 million for 500,000 records over five years) plus USD 5.0 million for updates Tempus 10-Q Q3 2026 2026-10-22 a maximum, earned as records are delivered sec.gov
Tempus AI and Bristol-Myers Squibb, Onco SOW USD 137.5 million aggregate fixed payments to 2031-12-31 Tempus 10-Q Q3 2026 2026-10-14 not an annual amount sec.gov
Kopinor and the National Library of Norway NOK 45,000,000 for 2026, index-adjusted from 2027 agreement s.17 2025-12-19 not per publisher, not paid by an AI developer kopinor.no
Tempus AI and Pathos AI (2025) USD 200 million over three years, incl. USD 50 million upfront (paid as of April 2025) Tempus 10-K FY2025 2026-02-24 not a per-record price sec.gov
Taylor & Francis (Informa) AI-related revenues of USD 75m+ in 2024, "non-recurring" Informa RNS, 2024 full-year results FY2024 revenue across its AI relationships, not one deal investegate.co.uk
Wiley, second GenAI project USD 21 million; Q1 FY2025 USD 16 million (release) against USD 17 million (presentation) Wiley release and presentation 2024-09-05 buyer not named sec.gov
Wiley, first GenAI project USD 23 million Wiley FY2024 earnings release 2024-06-13 buyer not named sec.gov
Reddit, arrangements of January 2024 USD 203.0 million aggregate contract value, two to three years; at least USD 66.4 million expected in 2024 Reddit Form S-1 2024-02-22 not per buyer; counterparties not named sec.gov

The Spirit amounts are bids, not a completed sale, and the outcome of the 2026-11-17 hearing is not recorded here (see the Spirit Airlines data sale).

Wiley's own documents give USD 16 million and USD 17 million for one quarter, and both stay. The rows behind these counts are in the tracker of AI data licensing deals with parties, dates and sources.

Case Counts by Jurisdiction and Rulings on Record

The case tracker holds 70 rows in 12 jurisdictions, 53 of them in the United States, and 7 rows record proceedings with no published docket. Each row is in the case tracker of AI copyright lawsuits over training data, as of 2026-11-09, statuses to 2026-11-16.

Rows counted 2026-11-17.

Jurisdiction Rows
United States 53
Germany 4
Brazil 2
Canada 2
Ireland 2
Denmark 1
European Union 1
France 1
India 1
Italy 1
South Korea 1
United Kingdom 1
Total 70

No worldwide total is given, because unofficial trackers disagree.

The Nine Training Rulings by Court and Date

Nine rulings on record decide a training question, three in the United States, three in Germany and one each in the United Kingdom, India and the European Union, and the Court of Justice item is an Advocate General opinion, not a judgment. As of 2026-11-09, Getty UK and ROSS checked to 2026-11-16.

Case Court and entry Date Outcome on the record Source
Bartz v. Anthropic N.D. Cal., Dkt. 231 2025-06-23 training fair use; scanned purchased books fair use; pirated copies not excused court PDF
Kadrey v. Meta N.D. Cal., Dkt. 598 2025-06-25 summary judgment for Meta on that record; not a holding that the use is lawful court PDF
GEMA v. OpenAI LG Muenchen I, 42 O 14139/24 2025-11-11 section 44b UrhG covers corpus copies, not memorization; not final press release
Getty v. Stability AI (UK) [2025] EWHC 2863 (Ch) 2025-11-04 model not an infringing copy; trade mark infringement in part; appeal permitted National Archives
Kneschke v. LAION OLG Hamburg, 5 U 104/24 2025-12-10 dataset copying justified; BGH I ZR 281/25 pronouncement set for 2026-12-17 BGH notice
GEMA v. Suno LG Muenchen I, 42 O 763/25 2026-07-31 US training copies not fair use; memorization infringes; not final press release
ANI v. OpenAI Delhi High Court, CS(COMM) 1028/2024 2026-07-24 training prima facie fair dealing, s.52(1)(a); interim injunction refused judgment PDF
Thomson Reuters v. ROSS 3d Cir. 25-2153 2026-09-29 not fair use for 2,243 headnotes; affirmed 3d Cir. opinion
Like Company v. Google CJEU C-250/25, AG opinion 2026-11-03 training questions inadmissible; obiter that training is as a rule permitted under the TDM exception; not a judgment CJEU press release

The two GEMA rulings are not final, and the Federal Court of Justice set its LAION pronouncement for 2026-12-17.

Settlement Amounts on Record

The Bartz v. Anthropic settlement is the largest amount in the case tracker, a USD 1.5 billion non-reversionary fund with final approval and judgment on 2026-07-20 (Dkt. 680), and it settles claims over past conduct, not a license. On 2026-09-02 (Dkt. 688) the USD 2,203.56 per Claimed Work is an initial figure.

As of 2026-11-09, with no total row.

Proceeding Amount as stated Entry and date What it is Source
Bartz v. Anthropic, 3:24-cv-05417-AMO (N.D. Cal.), as of 2026-11-09 USD 1.5 billion non-reversionary fund; 482,460 works; about USD 3,000 gross per work (approval order); about USD 2,203.56 per Claimed Work for 449,731 Claimed Works before Anthropic's final USD 450 million payment Dkt. 680, 2026-07-20; Dkt. 688, 2026-09-02 final settlement of past claims, not a license; USD 2,203.56 is an initial figure court PDF
McKinney v. Scale AI, S.F. Superior Court CGC-24-620481 USD 12,500,000 gross final approval 2026-10-30 contributor-pay case in the related proceedings, outside the 70 rows settlement site
Clearview AI, N.D. Ill. 1:21-cv-00135 an equity stake, no cash fund approved 2025-03-20; vacated by 7th Cir. No. 25-1673, Doc. 69, 2026-07-13 vacated and remanded; no replacement recorded court opinion

Program Counts by Kind

The program terms table compares 143 programs, 140 with terms recorded from the operators' own pages or filings and 3 whose pages could not be read on 2026-11-09, and it holds 13 pairs of values that contradict each other on the same operator's own pages.

Counted 2026-11-17 from the table of 2026-11-09.

Kind Programs
Company data program (CD) 28
Marketplace (MK) 24
Content licensing (CL) 21
Intermediary (IN) 15
Referral program (RF) 15
Contributor platform (HD) 9
RL environments (RL) 8
Custom capture (CC) 7
Dataset vendor (DV) 6
Crawler or training control (CR) 5
Not verified (NV) 3
Service terms or bounty (SV) 2
Total 143

An intermediary handles a sale and does not buy the data, so the 143 is not a count of buyers. No payout is summed here: each program's terms sit in the tracker of program payouts and terms compared, dated.

Estimates of the Market, Side by Side, and Where They Disagree

Five research publishers state eight estimates of the AI training dataset market for 2024 to 2026, from USD 2.81 billion (Straits Research, 2025) to USD 4.44 billion (Fortune Business Insights, 2026), and each is the publisher's claim, not a count of contracts. Checked 2026-11-09, with no average row.

Publisher Year Estimate Report as of Source
Fortune Business Insights 2025 USD 3.59 billion 2026-11-02 (checked 2026-11-17) fortunebusinessinsights.com
Fortune Business Insights 2026 USD 4.44 billion 2026-11-02 fortunebusinessinsights.com
Global Market Insights 2024 USD 3.2 billion 2025-05 gminsights.com
Grand View Research 2025 USD 3.2 billion 2026-06 grandviewresearch.com
Straits Research 2025 USD 2.81 billion 2026-08-24 straitsresearch.com
Straits Research 2026 USD 3.39 billion 2026-08-24 straitsresearch.com
The Business Research Company 2025 USD 3.19 billion 2026-09 thebusinessresearchcompany.com
The Business Research Company 2026 USD 3.87 billion 2026-09 thebusinessresearchcompany.com

Each publisher names the market "AI training dataset market", but none states what the figure counts. USD 3.2 billion is Global Market Insights' figure for 2024 and Grand View Research's for 2025, an observation and not an explanation. Grand View Research's separate data collection and labeling market is a different market and is not compared.

Forecasts and Growth Rates That Contradict Their Own Figures

For 2034 the forecasts run from USD 15.39 billion (Straits Research) to USD 23.18 billion (Fortune Business Insights), a difference of USD 7.79 billion, and two publishers state growth rates that conflict with their own figures. As of 2026-11-09, periods checked 2026-11-16.

What differs Figures as stated Difference or conflict
Estimates for 2025 USD 2.81 billion (Straits Research) to USD 3.59 billion (Fortune Business Insights); USD 3.19 billion (The Business Research Company) and USD 3.2 billion (Grand View Research) between USD 0.78 billion
Estimates for 2026 USD 3.39 billion (Straits), USD 3.87 billion (Business Research Company), USD 4.44 billion (Fortune) USD 1.05 billion
Forecasts for 2034 USD 15.39 billion (Straits), USD 16.3 billion (Global Market Insights), USD 23.18 billion (Fortune) USD 7.79 billion
Forecasts for other years USD 16.3 billion for 2033 (Grand View Research); USD 8.45 billion for 2030 (Business Research Company) horizons differ, not compared with 2034
Global Market Insights growth rate states 20.5 % for 2025 to 2034; USD 3.2 billion (2024) and USD 16.3 billion (2034) imply about 17.7 % a year over 2024 to 2034 stated and implied rates differ; periods one year apart
Business Research Company growth rate 21.6 % and 21.5 % on one page; 21.5 % for 2025 to 2026 and 2026 to 2030, 21.6 % for growth to 2030 two rates for one forecast

No average or midpoint is derived from these rows. The definitions and gaps behind each figure are on the page for AI training data market size.

Eight companies publish a running total of what they say they have paid, every total is the payer's own claim, and no seller, filing or court has confirmed an amount received under any of these programs. Pangea.ai is a marketplace and Shfa.ai a program with no published legal entity, so neither is called a buyer.

As of 2026-11-09, alphabetical, with no total, ranking or sum.

Company (kind) States it paid To whom, as its page puts it Period or date Source
Handshake AI (company data program) USD 250 million "partners" (undefined); "1K+ partner companies" undated; re-read 2026-11-10 joinhandshake.com
Labelbox Prism (company data program) USD 100M+ distributed not specified; "1k+ companies licensing" "to date"; checked 2026-11-09 prism.labelbox.com
Mercor, figures block USD 50,000,000 "companies for their data" undated; checked 2026-11-09 api.data.mercor.com feed
Mercor, stats block USD 22,000,000 (17,000,000 on 2026-11-09; 20,000,000 on 2026-11-10) paid to companies re-read 2026-11-16 api.data.mercor.com feed
micro1 (company data program) USD 5.8 million to 9 businesses, average over USD 600,000 businesses "the past 24 hours", LinkedIn post of 2026-09-10 (poster's own profile gives the title CEO) linkedin.com
Outlier (contributor platform), home page USD 500M+ experts and contributors (pay for work, not data sales) undated; checked 2026-11-09 outlier.ai
Outlier, about page USD 600M+ experts and contributors undated; checked 2026-11-09 outlier.ai
Pangea.ai (code marketplace, intermediary) USD 6 million earned by code owners; largest deal USD 300,000 (326 repositories) code owners "the last 60 days", undated; checked 2026-11-09 pangea.ai
Shfa.ai (medical data program) USD 300,000 to 3,000,000 paid to medical facilities, "representative and potential outcomes" medical facilities undated; re-read 2026-11-17 shfa.ai
Troveo, content-owners page USD 50M+ paid to creators creators no date; re-read 2026-11-10 troveo.ai
Troveo, business-data page USD 20M+ paid to licensors licensors no date; re-read 2026-11-10 troveo.ai

Mercor's stats block rose from USD 17,000,000 to USD 22,000,000 between 2026-11-09 and 2026-11-16 while its figures block states USD 50,000,000, and its referrer totals differ the same way, USD 90,000 against USD 2,000,000 "in 2026".

Outlier's two pages differ by USD 100 million and Troveo's by USD 30 million. Advertised and confirmed amounts are kept apart on the page for reported payouts to data sellers.

Figures Left Out of This Page, and Why

Several figures circulate in press and on competitor pages that no party, filing or court has confirmed, so this page names the topic and leaves the number out. The list is as of 2026-11-17 and carries no amounts:

  • press-only deal values for Google and Reddit, News Corp and OpenAI, and Amazon and the New York Times;
  • the buyer and term of the HarperCollins book deal;
  • Spirit's opening bid and a late micro1 proposal in the same sale;
  • a press-only valuation round of a human-data company;
  • competitor market-size snippets and competitor counts of public prices.

AI adoption, usage, compute and model-size statistics are outside this reference. Per-unit figures on the record sit with the page on AI training data prices.

What These Statistics Cannot Tell a Data Holder

A count or a stated amount describes what was recorded for other holders and other buyers, and it does not say what any one company's data would be offered.

Benefits

  • Dated anchors: each amount or estimate comes with its source and as-of date.
  • Visible contradictions: where two sources disagree, both figures stay in view.

Risks

  • A buyer's claim read as a payment: no seller has confirmed an amount received under any program.
  • A bid read as a sale: the Spirit amounts are bids, not approved as of 2026-11-09.
  • An estimate read as demand for one dataset: it is a publisher's model, not an offer.

Downsides

  • 55 of the 74 deals are marked not disclosed by any party.
  • No program payout is confirmed by a recipient.

Alternatives

  • Ask for a written offer and compare its terms with the stated amounts above.
  • Consider a non-exclusive license or a revenue share instead of a one-off fee.
  • Use an independent introduction to buyers that match the data.
  • Not selling or licensing the data at all, which stays a valid outcome.

What company data is worth to AI companies depends on rights and the buyer, and this page offers no valuation.

Where These Figures Are Detailed: Prices, Revenue, Funding, History and Terms

These statistics index the market section; prices, filed revenue, funding, history, size, payouts and terms each have their own page.

How to Cite a Figure From This Page

Cite the figure with its stating party, its as-of date and the table anchor, for example /market/statistics/#size-estimates for a market estimate. Name the publisher or filing as the party that states the number.

How Often These Statistics Are Re-checked

Tracker counts are recounted whenever a row is added, estimates yearly and when a publisher updates its page, filings quarterly and court dockets before each update.

Frequently Asked Questions

How many AI data licensing deals are on record?

The deal tracker holds 74 deals announced from 2022-10-25 to 2026-10-22, and 9 of them carry an amount stated by a party or a filing. The counts are as of 2026-11-09.

How big is the AI training data market?

Four publishers estimate the market for 2025 at USD 2.81 billion (Straits Research), USD 3.19 billion (The Business Research Company), USD 3.2 billion (Grand View Research) and USD 3.59 billion (Fortune Business Insights), each its own claim. Checked 2026-11-09, not averaged.

How many AI training data lawsuits are there?

The case tracker holds 70 rows in 12 jurisdictions as of 2026-11-09, and no worldwide total is given because unofficial trackers disagree. Nine rulings on record decide a training question.

How much have AI data buyers paid out?

Companies claim totals such as USD 250 million (Handshake AI) and USD 100M+ (Labelbox Prism), and no seller has confirmed an amount received under any program. Read between 2026-11-09 and 2026-11-16.

This page is information, not legal or financial advice.