Data Marketplaces & Intermediaries

AI's data-licensing market keeps paying publishers — and is shifting to pay-per-fetch

Source: Quartz · Sep 18, 2026

The market for licensed training data is maturing into a real, recurring line of business for content owners. OpenAI is now reported to hold around two dozen publisher and data-licensing agreements, the largest of them a roughly $250 million, five-year deal with News Corp, alongside arrangements with the Associated Press, Shutterstock (which also licenses to Meta), Condé Nast and Axel Springer. On the platform side, Reddit has disclosed roughly $203 million in aggregate licensing value from its own data deals.

The more interesting shift is structural. Early licensing looked like a flat annual fee for the right to train on a corpus. The emerging model is pay-on-fetch — compensation tied to retrieval and grounding, where a rights-holder gets paid when its content is actually pulled in to answer a query, not once up front. That change turns data licensing from a lump-sum sale into something closer to metered, ongoing revenue, and it rewards content that is repeatedly useful rather than merely large.

Estimates put the overall market around $3.59 billion in 2025 and climbing toward $4.44 billion in 2026. The signal for anyone sitting on proprietary data is that a legitimate, contracted, rights-cleared path to monetization now exists and is growing — and that provenance and the ability to prove where data came from, and that it was licensed with permission, is quietly becoming the thing buyers pay a premium for.

Key Points

  • OpenAI now holds roughly two dozen publisher and data-licensing deals; the largest is reported at ~$250M over five years with News Corp
  • Other counterparties include the Associated Press, Shutterstock (also licensing to Meta), Condé Nast and Axel Springer
  • Reddit has disclosed roughly $203M in aggregate licensing value from its data agreements
  • Deal structure is shifting from flat annual training fees toward pay-on-fetch / grounding models; the market is estimated at ~$3.59B in 2025 rising to ~$4.44B in 2026