AI Licensing Deals With News Publishers: Every Disclosed Price

News publishers have signed more AI licensing deals than any other kind of rights holder, and almost none of them carries a price anyone has stated on the record. The largest reported figures are News Corp's deal with OpenAI, which The Wall Street Journal said "could be" close to $250 million over five years, and Meta's deal with News Corp, reported at up to $50 million a year for at least three years. Below them sit Amazon and The New York Times at a reported $20 million to $25 million a year and OpenAI's minimum of around $16 million a year to Dotdash Meredith. Every one of those figures was reported by the press, not announced by the parties.

This guide is for anyone who needs the real numbers: a publisher weighing an offer, a company that holds a text archive and wants to know whether news prices apply to it, counsel drafting a licence, and reporters and investors tracking the market. It lists every news deal in our deal ledger that has a figure attached, sorted by amount and marked stated or reported; then the deals with no price; how the deals are structured (fixed fees, usage marketplaces, Google's pilot, revenue shares, payment in kind); the publishers that sued instead, and those that signed after suing; what the prices mean per year; the trend; and why none of this transfers to company records.

AIData.Broker keeps the deal ledger and case ledger used here and acts as the agent of US holders who license or sell their data, so every figure below links to its source for you to check. This guide is research from public sources, not legal, tax or investment advice.

Contents

1. Every publisher deal with a figure on record
2. The deals without a price
3. How the deals are structured
4. The publishers that sued instead
5. Signing after suing, and doing both at once
6. What the prices mean per year
7. The trend over time
8. Why news prices do not transfer to company records
9. Questions people ask
10. What this means for you

Every publisher deal with a figure on record

News publishers are the largest group in our ledger: close to half of the 111 disclosed AI licensing deals we track are with news organisations. Across all kinds of rights holder, only 19 deals carry a dollar amount at all, and for news the amounts that exist almost all come from the same two sources: reporters with unnamed sources, and analysts reading a company's financial documents. Our ledger marks each row stated (a party or a filing said it) or reported (the press said it, citing people familiar), and the difference matters. A reported figure may be a first-year amount, a ceiling ("up to"), or a number that bundles cash with technology credits.

The table lists every news deal with a figure attached, largest first by yearly amount. Where a source gives only a total, the yearly figure is the source's own arithmetic, not ours. A figure resting only on an aggregator's deal list is marked so; we keep it for completeness but never rely on it alone.

One more figure belongs beside the table, because it is quoted often and is not a deal. In late 2023, Apple was reported to be in talks with Condé Nast, NBC News and IAC over training licences, with reported offers of $50 million or more per publisher over several years. That figure comes from an aggregator's list, describes offers rather than signed contracts, and we found no announced Apple deal behind it.

What the table shows

Three things stand out. First, no news price in the table is stated by both parties. The closest to a primary source is Dotdash Meredith's minimum, which Adweek read from public financial documents of its parent IAC ( the record), and Perplexity's own pool, which describes a programme rather than a contract with one publisher. Second, the prices cluster by buyer and size: the largest publishers with the strongest archives got eight-figure yearly sums, national publishers in India were reported at single-digit millions, and licences to enterprise customers through Snowflake were reported at six figures. Third, the biggest numbers are for training and display together. The deals limited to showing content in answers, which are most of the ledger, have no price at all.

Our rates page turns the text deals with a yearly figure into one comparable: the median of public prices for text and archives on the sales that happened side is $16M per contract year. That median mixes news publishers with Reddit and the academic publishers, so read it as the middle of a small and unusual set, not as a price for any archive.

The deals without a price

Most news deals were announced with a press release, a quote from each chief executive and no money. The table groups them by licensee, because the licensee decides the shape of the deal more than the publisher does. Each linked name is one record in the deal ledger with its source and the source's own words.

Two patterns are visible without any prices. OpenAI signed first and most: it is the licensee in more news deals than any other company, and across all kinds of rights holder it has signed 29 of the deals in our ledger. And the use narrowed over time. OpenAI's early contracts with the Associated Press, Axel Springer, Le Monde, Prisa and the Financial Times all included training. The later wave, led by Meta, Google and Microsoft, licenses content to be shown, quoted and linked in answers, with training either excluded or not mentioned. Across the whole ledger, 40 deals license content only for display.

How the deals are structured

Behind the press releases there are only a handful of deal shapes, and a publisher (or any holder of text) negotiating today will be offered one of them. They differ in who carries the risk: a fixed fee puts it on the buyer, a usage fee or revenue share puts it on the publisher, and payment in kind puts it on the publisher's ability to turn technology into money.

Fixed annual fees

The headline deals are fixed fees for a multi-year term. VentureBeat's report of The Wall Street Journal's figure for News Corp and OpenAI spells out the arithmetic: " close to $250 million over five years, or $50 million a year." Meta's deal with News Corp is reported as "up to" a yearly sum for "at least three years" ( the record), so it carries a ceiling and a minimum term. Axel Springer's deal with OpenAI adds an upside: per the Financial Times, "Particularly popular content could generate additional revenue," and the deal "is not exclusive" ( the record). Fixed fees are what publishers want: predictable, independent of whether the AI product succeeds, and comparable year to year.

A floor plus a variable part

Dotdash Meredith's deal with OpenAI is the best documented hybrid. Adweek read in IAC's financial documents that OpenAI pays " a minimum of around $16 million per year," with a variable component on top to be calculated later. Our licence term on price structure takes it as the model to copy: anchor on a fixed minimum, and add refresh or usage fees ( the term).

Microsoft's marketplace

In February 2026 Microsoft opened its Publisher Content Marketplace, co-designed with publishers including the Associated Press, Condé Nast, Hearst Magazines, Vox Media and USA Today. Publishers "can also establish their own pricing models, choose specific AI companies as partners, and define usage parameters for their content," and "Microsoft takes a commission on transactions but has not disclosed the exact percentage" ( the record). Copilot is the first buyer. People Inc.'s chief executive described his company's part in it as pay-per-use ( the record). A marketplace shifts the risk to the publisher (no use, no money) in exchange for control over price and partners. It is also the first structure where an intermediary takes a fee from the publisher's side, which is worth knowing when comparing it with a direct deal.

Google's News AI pilot

Google's programme is the biggest by number of publishers. It began in December 2025 as a commercial pilot of AI article overviews on Google News, in which "publications participating in the commercial pilot program will receive direct payments from Google" ( the record). By its October 2026 update, Google said it "pay [s] participating news publications for the delivery and use of enhanced content to develop new AI features," and that more than 400 publications now take part, including Axios, CNN, the Financial Times and Forbes ( the record). No amounts or contract lengths are stated. Two individual pieces leaked: Reach, the UK publisher of the Mirror and the Express, called its deal time-limited, and Digiday understood it to run two years ( the record); CNN described its own as multiyear ( the record).

Perplexity's revenue shares

Perplexity has tried two shares. Its Publishers' Program, launched in July 2024 with partners including Automattic, Der Spiegel, Fortune, The Texas Tribune and TIME, shares advertising revenue when a publisher's content is cited, with the split undisclosed ( the record). In August 2025 it launched Comet Plus, a subscription, " setting aside $42.5 million for the first phase" and saying " 80% of Comet Plus revenue will go to participating publishers." Our lever on rights-holder splits uses that share as a benchmark beside book-licensing splits ( the lever). A revenue share is only worth what the product earns, and Perplexity is also a defendant in a string of news lawsuits (see below).

Payment in kind

A surprising number of news deals pay partly or wholly in technology. OpenAI funds Axios's expansion into new local markets ( the record); Village Media got funding, API credits and technical support ( the record); Grupo Folha and Grupo UOL got ChatGPT Enterprise, API and Codex access ( the record); Gannett got Perplexity's Sonar API and Enterprise Pro for all employees ( the record); the Guardian adopted ChatGPT Enterprise ( the record); and the Associated Press gets access to OpenAI's technology ( the record). Credits and tools have real value, but they are not revenue a publisher can bank, and they tie the publisher to the buyer's platform. When an offer includes them, value them separately from the cash.

Licences to companies, not AI labs

The newest structure skips the AI labs. Through Snowflake's Cortex Knowledge Extensions, seventeen publishers license real-time retrieval to companies, mostly financial institutions, for their own internal AI tools, mixing flat fees and usage fees paid from the customer's Snowflake credits, with reported deal sizes in six figures ( the record). This is the news version of a point that matters for every holder: the buyers of data are not only the handful of labs that train frontier models. Our buyer board maps the others.

The publishers that sued instead

While most publishers negotiated, a large group went to court, and the lawsuits shape every price in the first table. A buyer that faces a credible claim has a reason to pay; a buyer that expects to win on fair use (the US defence that allows some unlicensed copying, judged case by case) has a reason not to. Our case ledger tracks 83 lawsuits, rulings and settlements over training data; the table lists the ones brought by news and reference publishers.

Where the law stands

None of the US news cases has reached a ruling on fair use, the question that decides them. The New York Times' complaint seeks "billions of dollars in statutory and actual damages" ( the record); summary judgment motions are pending. In September 2026 the US Justice Department filed a non-binding statement in the consolidated OpenAI cases urging the court "to hold that using copyrighted written works to train large language models (LLMs) is fair use" ( the record); the court has not ruled. Outside the US, the Delhi High Court refused ANI an interim injunction with a provisional view that storing its works fell within an exception in India's copyright law ( the record), and a Hungarian publisher's case against Google has put the first AI copyright questions to the EU's Court of Justice ( the record).

The rulings that do exist come from books and legal research material, not news, and they cut both ways. Our guide to fair use and AI training walks through them. The one settled number is from books too: the Anthropic class settlement pays about $3,000 per work out of a $1.5 billion fund ( the record). Every news licence price is negotiated in the shadow of those open questions.

Suing the middlemen

Two cases target the supply chain rather than the model builder. News Corp sued Brave, alleging its search engine disguises its crawlers to evade publisher blocks and delivers "verbatim or near verbatim" news summaries to AI companies ( the record). Reddit's suit against Perplexity and three scraping firms runs on the same theory for user posts (see our Reddit guide). And in the UK, Reach told Digiday it had sent OpenAI a cease-and-desist letter "as it steps up pressure on AI firms using its content without a deal," in the same month it signed with Google ( the record).

Signing after suing, and doing both at once

The line between licensors and litigants is not clean. Several publishers sue one AI company while licensing to another, and at least two signed with the company they had sued. That is not inconsistency; it is how a market price forms when the law is unsettled. A lawsuit against one buyer raises the cost of taking for all buyers, and a licence with another sets a reference price that a court can later look at.

The plaintiffs in the publishers' suit against Cohere, as named in the report our record cites, include several that license elsewhere, among them Condé Nast, The Atlantic, Vox Media and The Guardian ( the record). Music shows the end state of the same pattern: Warner Music Group sued Suno and then settled into a licensing partnership ( the record). For a holder the lesson is practical: being willing to enforce does not close the door to licensing, and a licence can be the way a dispute ends.

What the prices mean per year

The reported figures are easiest to compare per year, and on that basis they fall into tiers. At the top, News Corp's two deals each come to about $50 million a year with OpenAI and up to $50 million a year with Meta. Next, The New York Times with Amazon at $20 million to $25 million a year, Dotdash Meredith with OpenAI at a $16 million minimum, and Axel Springer at "tens of millions of euros". Then national publishers outside the US and UK, at around $5 million and about $3 million a year. At the bottom of the reported range, enterprise retrieval licences through Snowflake in six figures.

What a buyer is paying for

These are not prices for words. A news licence bundles at least four things, and the mix explains the tiers better than archive size does:

1. A live feed. Most deals deliver today's journalism as it is published, which an AI product needs to answer questions about current events. That is a service, priced per year.
2. Display rights with a brand. The right to quote, summarise and link to a named, trusted source inside an answer. The 40 display-only deals in our ledger are this alone.
3. Training rights to an archive. Decades of edited, fact-checked text. This was the core of the 2023 and 2024 deals and has largely dropped out of the later ones.
4. Peace. A licence removes, or at least reduces, the risk of being sued by that publisher. The tier list lines up with litigation power: the publishers paid most are the ones with the resources and the record to sue, and News Corp and The New York Times do both.

Comparisons that do work

The fairest comparisons for news prices are other licences of large, living text collections. Reddit's January 2024 licences carry $203.0 million of contract value over two to three years, stated in its IPO filings ( the record). Academic publishers sell archives with less of the feed and more of the one-off: Wiley reported AI licence revenue of $49.1 million in the year to April 2026 ( the record), and Informa said its AI partnerships produced $75m+ of non-recurring data access revenue for Taylor & Francis in 2024 ( the record). Book publishers price per title: HarperCollins' three-year deal paid $5,000 per title, split evenly with authors ( the record).

The trend over time

News deals came in waves, and the waves changed shape. The chart counts every news publisher licence in our ledger by the half-year it was announced, split by whether it includes training, and shows how many carry a dollar figure (stated or reported).

Sources: every news publisher licence in the deal ledger, each listed with its record in the two tables above, counted by the half-year of its date. Deals whose use the source did not state are counted in the total only. The last half-year runs to 17 October 2026.

The chart tells a clear story. The first wave, from mid-2023 to mid-2024, was OpenAI buying training rights from the largest publishers it could sign, with the biggest reported prices. The second wave, peaking in the second half of 2025 with Meta's batch of seven publishers in one announcement and Google's pilot, was platforms buying display rights, with almost no prices. Since then, deals have kept coming at a steady pace, the training share has fallen to nearly nothing, and the few prices reported are smaller (the Indian deals) or come with a ceiling (Meta's "up to"). Meanwhile the lawsuits multiplied: most of the news cases in the table above were filed in 2025 and 2026.

One reading is that labs collected the archives they needed for training early and now pay for freshness and display. Another is that unsettled fair use law lets them avoid paying for training at all until a court decides. Both point the same way for a holder of text: the money now is in what is new and in the right to show it, not in an archive alone.

Why news prices do not transfer to company records

Holders of business data often read the News Corp figure and assume their archive is worth something similar. It is not, and the reasons are structural, not a matter of size.

The biggest single sale of company records on record shows the difference in scale and in kind. Google's winning bid for the whole corporate archive of bankrupt Spirit Airlines was $10M, once, for records that will never be shown to anyone ( the record). A news publisher can earn more than that in a year because it sells something renewable and public. A company's records are valuable for the opposite reason: they show how real work is done, which nothing on the open web shows. Our guides to how much company data is worth and selling company data to AI labs explain how those prices are set, and the price board lists them.

Questions people ask

How much does OpenAI pay news publishers?

OpenAI has not disclosed a price for any news deal. The reported figures range from about $50 million a year for News Corp (from a reported $250 million over five years) down to about $3 million a year for the Indian Express Group. Many of its later deals include technology access or funding rather than, or beside, cash ( Axios, Folha and UOL).

What is the biggest AI licensing deal with a news publisher?

By reported value, News Corp's five-year deal with OpenAI, which The Wall Street Journal said "could be" close to $250 million ( the record). Meta's deal with News Corp, reported at up to $50 million a year for at least three years, is of similar yearly size. Neither figure was confirmed by the parties.

How much does Amazon pay The New York Times?

The Wall Street Journal reported $20 million to $25 million a year for news, NYT Cooking and The Athletic content, including training Amazon's models; neither company disclosed the terms ( the record). The Times is at the same time suing Microsoft, OpenAI and Perplexity.

Does Google pay publishers for AI?

Yes, through its News AI pilot, in which Google says it pays participating publications "for the delivery and use of enhanced content to develop new AI features"; it named more than 400 publications in October 2026 ( the record). No amounts are public. Google also licenses a real-time feed from the Associated Press for Gemini ( the record).

Why did The New York Times sue OpenAI instead of licensing?

The Times alleges Microsoft and OpenAI copied its journalism without permission to build their products, and seeks "billions of dollars" in damages ( the record). Its core copyright claims survived a motion to dismiss, and summary judgment motions are pending with no trial date ( the record). It licensed to Amazon in 2025, so it is not against licensing as such.

What is Microsoft's Publisher Content Marketplace?

A marketplace Microsoft opened in February 2026 where publishers set their own prices and usage terms for content used to ground AI answers, with Copilot as the first buyer and Microsoft keeping an undisclosed share ( the record).

Are news AI deals for training or for showing content in answers?

Increasingly for showing. The early OpenAI deals included training on archives; most deals since mid-2024, led by Meta, Google and Microsoft, license content to be summarised, quoted and linked in answers. Across our ledger, 40 deals are display-only against 56 that include training, and the training share among news deals fell from half in the first half of 2024 to none so far in the second half of 2026.

What this means for you

If you are a publisher, the public record gives you anchors, structures and leverage, but no price list. If you hold any other kind of text, it tells you what buyers pay for (freshness, display, rights, peace) and warns you off borrowing a news figure for an asset that is not news.

1. Separate the four things you could sell: a feed, display rights, training rights to an archive, and a release of claims. Price each one; do not let a buyer bundle them into one number.
2. Insist on a floor. A fixed minimum, as in Dotdash Meredith's deal, protects you from a product that never earns; usage fees and revenue shares can sit on top. Our licence terms set out where to stand on each clause, and what is a data licensing agreement explains the contract.
3. Value technology credits separately from cash, and check whether they lock you to the buyer.
4. Keep the option to enforce. Licensing to one buyer and suing another is normal in this market; see is it legal to scrape data for AI and the case ledger.
5. Use the right comparables. For company records, start from our valuation, the price board and the rates, not from News Corp. If you already have an offer, the offer check compares it with the record.

AIData.Broker acts only for holders: the holder licenses or sells its own data to the buyer, and pays us a fee fixed in the mandate only if a sale closes; we run sales for US holders. Whether to go direct or through us is your decision. For the rest of the market, see our Reddit guide, how big the AI training data market is, how AI companies get training data, and AI training data prices. Every figure here is on the records page with its source; how we source every number explains the method.