Since 2023, Shutterstock has signed multi-year data licensing agreements with OpenAI, Meta, Google, Amazon, and LG AI Research, feeding its library of hundreds of millions of images, videos, and 3D assets into the training pipelines of some of the industry's largest foundation models. According to AI Business, the company continues to expand this strategy, treating licensing deals with AI developers as a core, recurring line of business rather than a one-off cash grab.
That shift matters because Shutterstock's original business — selling individual image and video licenses to marketers, publishers, and designers — has been squeezed from two directions at once. Generative image tools let customers produce a usable visual in seconds instead of searching a stock library, cutting into per-download revenue. At the same time, the same generative tools need enormous volumes of licensed, rights-cleared visual content to train on, and few companies can supply that at Shutterstock's scale with a defensible legal position.
Shutterstock's answer has been to sell into the disruption rather than only compete against it: license the raw material to the model builders while continuing to sell finished images to everyone else.
Why stock libraries became AI training infrastructure
Large image and video models need paired data — a picture and a caption, a clip and a description — at volumes that scraping the open web can't reliably deliver without running into copyright disputes. Stock libraries already have this pairing built in: keywords, captions, and metadata attached to every asset for search purposes double as ready-made training labels. Shutterstock has been positioning that structure as the product, not the individual photograph.
- Pre-tagged, searchable metadata that doubles as training labels
- Rights clearance the company can indemnify against, unlike scraped web images
- Continuous new supply from its base of contributors, keeping datasets current
The compensation question
The other half of this business model is Shutterstock's contributor base — the photographers and videographers whose work fills the library. Shutterstock pays contributors through its Contributor Fund when their content is used in AI training datasets, a mechanism the company points to when defending the licensing strategy publicly. Whether that compensation is proportionate to what AI labs actually pay Shutterstock for the data remains a point of friction among contributors, and Shutterstock has not published a full breakdown of how licensing revenue is split.
What this means for teams building with AI
For engineering and product teams working with generative image or video models, Shutterstock's expanding deal roster is a useful signal, not just a business story.
- Provenance is becoming a selling point. Models trained on licensed, rights-cleared data give enterprise customers a cleaner legal footing than models trained on unclear or scraped sources — a differentiator worth checking before you commit to a vendor for commercial output.
- Licensed-data models may lag on some benchmarks. Curated, licensed datasets are smaller and narrower than the open web; teams should expect trade-offs between legal safety and raw capability, and should benchmark rather than assume.
- The supply chain behind a model matters for procurement. As more foundation model vendors disclose data-licensing partners, that list is becoming a legitimate line item in vendor risk assessments, alongside uptime and pricing.
AiiN's takeaway
Shutterstock's pivot is a reminder that the generative AI boom created two markets at once: one for AI-generated output, and one for the licensed data needed to produce it responsibly. Shutterstock chose to sell into the second market instead of only defending the first, and its growing list of AI licensing partners suggests that bet is paying off financially even as its traditional stock-photo business faces pressure. In our estimation, more content libraries — stock media, but also niche archives in audio, video, and specialized imagery — will likely follow the same playbook: license to the model builders, keep selling to everyone else, and let provenance become the pitch.