The Cost of Data: Why AI Companies are Paying for Content in 2026

The Pulse

There were 12 publicly disclosed AI content licensing deals in 2023. By the end of 2025 that figure had reached 91. By mid-2026 it is projected to hit 127, and nearly all of the growth is concentrated in a category that barely existed three years ago, live retrieval and attribution deals rather than one-time training data dumps.

The shift from training to live access is the real story here. Deals tracked for ongoing feeds, grounding, and real-time attribution went from 2 in 2023 to 11 in 2024 to 18 in 2025, with 34 projected for 2026 alone.[Media and the Machine AI content licensing deals June 2026]

AI companies are no longer just buying historical archives to train models once. They are buying ongoing pipelines of fresh content their models can pull from in real time, every time a user asks a question.

Core significance

Why it matters:

• OpenAI is the clear leader in deal volume, almost double its nearest competitor:  OpenAI has 24 publicly announced licensing agreements, nearly twice Microsoft and Meta’s totals, a pattern that suggests the company sees proprietary content access as a strategic moat rather than purely legal protection.
• The New York Times, having sued OpenAI in late 2023, has since struck a separate paid deal with Amazon worth an estimated 20 to 25 million dollars annually:  Copyright lawsuits and licensing deals are now running in parallel rather than as alternatives, with some publishers pursuing both tracks simultaneously against different AI companies. [AI Business AI lawsuits 2026 settlements licensing litigation]

Getty Images, one of the original plaintiffs that sued Stability AI for copyright infringement back in 2023, has since built a separate licensing partnership with Nvidia, turning what began as litigation into a parallel revenue line.

• Publisher leverage varies enormously, and no single outlet controls the market:  Even the largest individual publishers hold only single-digit deal counts, Shutterstock 7, Wikimedia 6, Reddit 5, while a long tail of smaller publishers collectively accounts for far more deals combined.[Digiday publishers scorecard Big Tech AI licensing deals]

Deep Context: From quiet payouts to a structured marketplace

The earliest licensing deals were small and largely defensive. The Associated Press signed with OpenAI in July 2023 for an undisclosed amount, becoming the first major news organization to do so. Axel Springer followed in December 2023 at roughly 13 million dollars a year over three years, covering Business Insider, Politico, Bild, and Welt. [AI Watch Dog content licensing deals tracker]

Reddit’s February 2024 deal with Google, struck the same day Reddit filed for its IPO, set a new benchmark at roughly 60 million dollars a year for real-time access to Reddit’s forums. OpenAI followed months later with its own Reddit deal, reportedly worth around 70 million dollars annually.

That pricing escalation has continued. News Corp’s current arrangement with OpenAI is worth approximately 250 million dollars over five years, covering the Wall Street Journal, the Times of London, the New York Post, and the Australian. The company holds a separate deal with Meta worth up to 50 million dollars per year, meaning one media portfolio now commands meaningful fees from two of the largest AI platforms in the world simultaneously.[Tenpoint Labs AI publisher licensing deals brand visibility]

The 1.5 billion dollar settlement that reset the price floor

Anthropic’s September 2025 settlement of a class-action copyright lawsuit, the largest publicly reported copyright recovery in history at 1.5 billion dollars, did more than resolve one company’s legal exposure. It established a 3,000 dollar per work baseline for copyright valuation in AI training contexts, giving every publisher and author a concrete number to negotiate from going forward.[Will Scott AI licensing deals search visibility 2025]

As covered in our OpenAI vs Anthropic report, the two companies have taken visibly different public postures toward content owners, and the settlement’s baseline pricing is now functioning as an industry reference point well beyond the case that produced it.

Data Insights

By the numbers:

All figures from named publisher trackers, court filings, and company disclosures cited inline.

• News Corp’s Meta deal alone is worth up to 50 million dollars a year over three years, roughly 150 million dollars total:  The Financial Times’ OpenAI deal runs an estimated 5 to 10 million dollars annually, while Vox Media’s multi-year OpenAI agreement covers Vox, The Verge, New York Magazine, Eater, and SB Nation under a single undisclosed sum.[Everything PR billion dollar bailout publisher licensing tracker]
• Amazon claims more than 200 publisher deals, though independent trackers count closer to 13 confirmed partnerships:  That gap between claimed and independently verified deal counts is common across the market, since most contract values and even existence are not publicly disclosed in full.
• Total disclosed licensing deal value across the market is approaching roughly 2 billion dollars combined:  That figure spans dozens of individual agreements ranging from undisclosed five-figure arrangements with niche publishers up to the 250 million dollar News Corp and OpenAI deal at the top end.

Table 1: Major AI content licensing deals compared

Publisher	AI company	Reported value	Deal type	Year signed
News Corp	OpenAI	250 million over 5 years	Training plus attribution	2024
News Corp	Meta	Up to 50 million per year	Training plus chatbot access	2026
Reddit	Google	60 million per year	Live retrieval, real time	2024
Reddit	OpenAI	70 million per year, estimated	Live retrieval, real time	2024
New York Times	Amazon	20 to 25 million per year	Training plus content access	2025

Table 2: How the licensing market has scaled since 2023

Year	Total disclosed deals	Live retrieval and attribution deals
2023	12	2
2025	91	18
2026, projected	127	34

IMAGE PROMPT BOXPrompt: A clean ascending bar chart infographic showing AI content licensing deal volume climbing from a small bar in 2023 to a much taller bar projected for 2026, using navy blue gradient bars on white background, minimalist financial chart style, no readable axis numbers needed.Alt text: AI content licensing deals growth 2023 to 2026 showing publisher payment trendPlacement: After Table 2, before The business case section

The Business Case: What the licensing shift means for publishers and AI buyers

For publishers, the strategic lesson from three years of deal data is that fragmentation, not consolidation, has defined market leverage so far. No single content owner controls enough share to set market price alone, which is precisely why the Really Simple Licensing initiative, backed by Reddit, Yahoo, Medium, and People Inc., is attempting to standardize terms across the industry the way ASCAP and BMI did for music.[CJR Reddit winning AI game licensing RSL]

For enterprises building products on top of major AI models, the practical implication is less obvious but increasingly relevant. Which publishers a given AI platform has licensing deals with materially shapes which brands and sources that platform’s outputs cite, meaning AI visibility is becoming a function of the underlying platform’s content partnerships, not just a company’s own SEO or content strategy.

As covered in our Anthropic IPO report, content licensing costs and copyright settlement exposure are now material enough line items that they factor directly into how investors are valuing frontier AI companies ahead of public listings, not just a legal compliance cost buried in the footnotes.

Enterprises evaluating which AI platform to standardize on for internal knowledge work and customer-facing products should treat licensing posture as a genuine differentiator, not a footnote. A platform with deep, well-compensated publisher relationships is less likely to face a sudden content blackout or injunction mid-deployment than one relying primarily on contested fair-use arguments still working through the courts.

Expert Nuance: Regulators may make licensing mandatory rather than voluntary

Every deal covered so far has been a bilateral, voluntary negotiation. A parallel regulatory track could change that. Policymakers in Europe, Indonesia, Latin America, and at the World Intellectual Property Organization are actively exploring statutory licensing regimes that would require AI companies to pay publishers automatically, for both past and future content use, rather than negotiating deal by deal.[Poynter global push AI companies pay for news statutory licensing]

Danielle Coffey, president of the News Media Alliance, has framed the dynamic bluntly: if publishers start winning their copyright lawsuits, an expectation of payment becomes standard, and the resulting prices are likely to be set higher than any government-mandated rate would have been. That creates an unusual incentive structure, where AI companies negotiating now, before more lawsuits resolve, may be locking in better terms than they would get later.

Europe’s longer tradition of licensing regimes and collection agencies makes statutory licensing more immediately implementable there than in the United States, where individual litigation outcomes are more likely to set the market’s effective price floor first.

Strategic Outlook

1. Watch whether RSL achieves real adoption or remains symbolic:  No major AI company has yet committed to honoring the Really Simple Licensing standard, even as the publisher coalition backing it grows. Reddit’s own spokesperson noted the company backs RSL despite already having an estimated 60 million dollar annual Google deal, suggesting even well-compensated publishers see value in standardized, industry-wide pricing.
2. Expect deal structures to keep shifting from one-time training payments toward recurring access fees:  The trajectory from 2 retrieval deals in 2023 to a projected 34 in 2026 suggests AI companies increasingly view fresh, current content as a recurring operational input, similar to a data feed subscription, rather than a one-time asset purchase.
3. Major entertainment studios remain the clearest holdout from the licensing trend, favoring litigation instead:  Disney, Warner Bros, and Universal have largely avoided licensing deals in favor of lawsuits, with Warner Bros Discovery’s September 2025 suit against Midjourney over Superman, Batman, and other characters as the clearest example of that alternate path playing out in parallel to the licensing market covered here.

Key Question Answered

Why are AI companies paying for content, and how much are they spending?

AI companies are paying publishers to reduce copyright litigation risk and to secure ongoing access to fresh, attributable content their models can cite in real time, not just historical archives for one-time training. The market has scaled from 12 disclosed deals in 2023 to a projected 127 by mid-2026.

Individual deal values range from undisclosed five-figure arrangements with smaller publishers up to News Corp’s 250 million dollar, five-year agreement with OpenAI, with Reddit’s roughly 60 to 70 million dollar annual deals with Google and OpenAI as the clearest example of the market’s recent shift toward paying for live, real-time access rather than static training data. Anthropic’s 1.5 billion dollar copyright settlement in September 2025 established a 3,000 dollar per work baseline that is now functioning as an informal price floor across the broader market, even for companies not party to that specific case.

The Takeaway

The AI content licensing market has moved past the experimental phase faster than almost any other part of the AI economy. What began in 2023 as a handful of defensive, undisclosed payments to avoid lawsuits has become, in three years, a structured market with public price benchmarks, standardization efforts, and deal volume doubling roughly every year.

The more important shift is not the dollar figures themselves but what they are buying. Early deals purchased a one-time license to train on an archive. Current deals increasingly purchase an ongoing relationship, live data feeds, real-time attribution, and a say in how a publisher’s brand surfaces inside an AI platform’s answers. That distinction matters because it changes the financial relationship from a one-time transaction into a recurring, renegotiable dependency, for both sides.

For publishers still holding out, the data suggests the choice is no longer simply license or litigate. It is choosing which of those two paths sets a better price, and the Anthropic settlement’s 3,000 dollar baseline means litigation now comes with a number attached that did not exist three years ago.