AI Data Licensing Deals Shift From One-Time Training Dumps to Real-Time Access
A new analysis tracking 91 publicly disclosed AI content and data licensing deals through June 2026 found that agreements involving attribution and real-time access grew from just 2 in 2023 to 11 in 2024, 18 in 2025, and a projected 34 in 2026 — a structural shift in what AI companies are actually buying, according to research from media analyst Rob Kelly and a related report from Pebblous.
News and journalism licensing dominates the disclosed market, accounting for 48 of the tracked deals, far ahead of music and audio (16) and images and video (12). Reported deal values vary enormously: News Corp's licensing arrangement reportedly averages about $50 million per year across its portfolio of titles, while smaller publisher deals with OpenAI have ranged from roughly $5 million to $16 million per year, according to reporting cited by Quartz.
Analysts point to two forces driving the shift toward real-time licensing: increased crawler-blocking by publishers narrowing what's freely scrapable, and the reality that models answering current-events questions perform meaningfully better with retrieval-augmented, continuously updated sources than with static, aging training data.
For individual consumers, a much smaller and newer parallel market has emerged in the form of data co-ops, which pool personal data from many contributors for collective licensing — though individual payouts in these programs remain modest, commonly just a few dollars, reflecting how early this specific segment still is compared to institutional publisher deals.
Learn more in our explainer on how AI companies buy and license data and how data co-ops work for individuals.