The Top Direct Suppliers of Licensed AI Training Data (2026)

AI models are only as good, and as ethically defensible, as the data they train on. For years, the default was to scrape the open web and answer questions later. But a wave of copyright lawsuits, new disclosure rules, and rising scrutiny have prompted AI labs to turn to ethically sourced data in which rights holders are fairly compensated. In response, a real market has formed: content owners, who are rights holders, are now licensing their libraries directly to AI buyers.

In this emerging market, “middlemen” have emerged in the distribution supply chain. Acting as brokers for rights-holders, they resell training rights to AI Labs on a revenue-share basis. This brokerage function serves a purpose in the supply chain for owners of small libraries who otherwise would not have access to buyers. The challenge is in assuring provenance in a fractured supply chain. Who actually has the rights, where did those rights come from, and where has the data already been sent for training? The chain of custody is lost. Leaving buyers with risk and sellers with uncertainty about what was paid for the rights and by whom. This system doesn't scale and lacks transparency. 

Large libraries that are themselves rights holders are a safer direct source for AI Labs.

We've compiled a list of the market leaders directly licensing video, image, and text to the AI hyperscalers. These companies own large, expansive libraries, and the content itself is high-quality, professionally produced, and highly usable for AI training. They've also spent years building direct relationships with buyers, learning how AI Labs think about data, what their requirements actually are, and how those requirements evolve.

In a market where legal risk, provenance, and compliance are still being worked out, buyers gravitate toward partners they feel confident in. These suppliers have earned that trust by being consistent, transparent, and reliable in how they operate. 

Here are the primary training data suppliers worth knowing.

CuriosityStream

Owns: a deep library of factual and documentary programming.

CuriosityStream’s AI-licensing business is now forecasting up to $80 million in 2026 revenue. Factual content, clear narratives, clean visuals and strong production standards. When AI teams are training models, consistency and quality of data directly impact outcomes. Not all video is equally valuable, and CuriosityStream’s catalog aligns well with what these models need. With Versos, CuriosityStream has also invested in the infrastructure for indexing and analysis, frame and scene-level metadata generation, precise dataset assembly based on buyer requirements and packaging and delivery direct to AI model training environments.

Best for: authoritative, factual and documentary video with clean title, delivered AI-ready with frame- and scene-level metadata and datasets assembled to buyer spec, so it drops straight into a training pipeline.

‍

Getty Images

Owns: one of the largest licensed visual libraries in the world, plus proprietary editorial and archival video.

Getty runs a full AI data-licensing business, negotiates bespoke dataset agreements, and compensates contributors whose work is used. VentureBeat called one Getty release the "cleanest" visual dataset for training foundation models. Licensing-first DNA, contributors paid, provenance documented.

Best for: rights-cleared, indemnifiable visual data.

‍

Newsflare

Owns: a licensed archive of user-generated and professionally shot real-world video.

Newsflare spent a decade licensing UGC video to broadcasters and now applies that rights-management expertise to AI training. Payment is simple and transparent: contributors earn $10–$50 per hour of content used and can opt out, keeping both consent and compensation clean.

Best for: authentic, real-world video rather than polished stock.

Shutterstock

Owns: hundreds of millions of images, plus video, music, and 3D assets (including the Pond5 footage marketplace it acquired).

Shutterstock is the clearest proof that clean, direct licensing is a real business: its AI-data-licensing arm has generated $100M+ on the strength of a six-year OpenAI agreement and deals reportedly worth $25–$50 million each with Meta, Google, Amazon, and Apple. One licensed counterparty, contributor payments built in, multimodal coverage.

Best for: large-scale, multimodal visual data from a single licensed source.
‍

GoPro

Owns: an enormous archive of first-person, POV action, adventure, sports, drone, and travel footage; real-world, high-variance video that's hard to source anywhere else.

GoPro launched an opt-in AI Training Licensing Program in 2025, and by late 2025, subscribers had contributed over 300,000 hours to it. Because GoPro owns the platform and licenses directly, the provenance is clean, and the payment is unusually simple: it runs a 50/50 revenue share with the subscribers who opt in their footage. One owner, explicit consent, transparent split.

Best for: authentic first-person and action video with a clear opt-in consent trail.

‍

Cineverse

Owns: a catalog of 66,000+ premium film and TV titles (including genre and anime) plus a proprietary metadata dataset of 2M+ titles.

Cineverse built Matchpoint "Reel Visuals AI," a purpose-built service that licenses its owned catalog to AI training buyers on a non-exclusive revenue-share basis. It owns the underlying content, which keeps provenance clean; note that Matchpoint also operates as a light technology layer for other content owners, so confirm which titles are Cineverse-owned in any given deal.

Best for: premium film/TV footage plus rich, structured metadata.

‍

U.S. Soccer

Owns: decades of match and event footage; full matches, player interviews, pre-clipped highlights, and multiple camera angles.

As the federation that owns its archive, U.S. Soccer is a textbook clean rights holder. It renewed an agreement in early 2026 to monetize and license that footage (managed through Veritone's licensing platform). For multi-angle, structured sports video, going to the federation that owns it is about as clean as provenance gets.

Best for: premium, multi-angle sports video with a single rights-owning holder.

‍
The through-line across every supplier here: the cleanest provenance and the simplest payment both come from going direct to the owner. In 2026, that paper trail is the product. Versos was built for exactly this. Our platform prepares and licenses video and multimodal training data with a documented chain of custody from source to model, so the data arrives rights-cleared, consent-checked, and defensible.

But, buying the rights is only half the job. Most data suppliers hand you enormous unstructured datasets your team still has to clean, tag, segment, and format before a model can touch them.

Versos closes that gap. Our platform delivers structured, model-ready datasets; clipped and segmented to your exact training requirements. It indexes and analyzes the content, generates frame- and scene-level metadata, segments and assembles datasets to your exact specifications, and delivers them straight into your training environment. The cleanest provenance, the simplest payment, and data that is model-ready on arrival.

If you are building models and need structured training data you can prove you have the right to use, let's talk.

‍