AI Training Data Demand Tracker: Where AI Labs Spend

Everyone agrees AI training data is valuable. Almost nobody can tell you which data, how valuable, or who is actually paying for it. Most licensing contracts are private, pricing is opaque, and the headlines mix real deals with lawsuits and speculation.

So we built a way to see it. The Versos AI Training Data Demand Tracker is a free, live read of where AI training data demand and money are concentrating across video, text and audio. It scores 10 data categories on public evidence and refreshes as new signals land.

Explore the AI Training Data Demand Tracker

What is the AI Training Data Demand Tracker?

The tracker monitors public reporting on model launches, licensing deals, investments and court filings, then turns that reporting into a ranked demand index for each category of training data.

• Public evidence only. Every score and figure traces to a named public source. No Versos customer or private data informs any score.
• Three modalities. Video, text and audio are tracked side by side, so you can compare demand across the whole AI training data market, not one slice of it.
• Evidence gaps stay visible. Where no public pricing exists, the category is flagged rather than guessed.

How we measure AI training data demand

Each category is scored from 1 to 5 on three forces:

1. Capability heat: momentum in the models and products that consume this data.
2. Scarcity: how hard the data is to get legally, at quality, at scale.
3. Proven spend: money actually disclosed through deals, investments and price signals.

The three scores roll up into a demand index from 0 to 100, which ranks the matrix from hottest to coldest. The pipeline runs in four steps: monitor, score, rank, refresh.

Which AI training data is most in demand right now?

Here is the demand matrix as of 9 November 2026. Scores are shown as capability heat / scarcity / proven spend.

Video training data: premium IP leads, generic stock trails

• Scripted narrative and film/TV: demand index 95 (4 / 5 / 5)
• Real-world and world models: demand index 85 (5 / 4 / 4)
• Egocentric and robotics: demand index 70 (4 / 3 / 4)
• Stock and B-roll: demand index 34 (2 / 1 / 2)

Scripted film and TV is the rights-clean, exclusive footage frontier labs want most and cannot scrape. The signals are concrete. Disney's agreement bringing its characters to OpenAI's Sora came paired with a reported investment of roughly $1B. Google is reportedly courting Hollywood studios for AI video rights at around $40M per copyrighted character. The MPA signed an AI copyright pact with ByteDance covering IP in its Seedance video models.

Real-world footage for world models and egocentric robotics data follow close behind. Mecka raised a $60M Series B to supply human-motion data to robotics labs, and Baidu launched Apollo World, a world model that simulates interactive road environments to train robotaxis.

Stock and B-roll sits at the bottom. Generic footage is abundant and largely commoditized. In video, the gap between content and training-grade content is wider than in any other modality.

Text training data: community and news remain the foundation

• Forums and community (UGC): demand index 88 (4 / 5 / 5)
• News and publishing: demand index 80 (3 / 4 / 5)
• Code and technical: demand index 58 (5 / 3 / evidence gap)

Text is where the most money has been disclosed. Reddit's licensing run-rate is reported at $60M to $70M a year across Google and OpenAI. News Corp's deal with OpenAI is worth $250M+ over five years, the publisher benchmark. In November 2026, Anthropic signed its first paid content and data licensing deal with a news publisher, the Associated Press, with terms undisclosed. Code scores high on capability heat, but no meaningful public pricing has surfaced yet.

Audio training data: licensed music is hot, and litigation is setting the price

• Music (licensed): demand index 82 (4 / 5 / 4)
• Speech and voice: demand index 60 (4 / 3 / 3)
• Sound and ambient: demand index 32 (2 / 2 / evidence gap)

In audio, much of the price discovery is happening in court. Reservoir and Shapiro Bernstein are seeking roughly $340M combined from Suno and Anthropic. The Elvis Presley estate is seeking about $41M from Anthropic at $150K per track, and Eminem's publisher Eight Mile Style is seeking $40M+ from Suno over 269 songs. These are amounts claimed, not awarded, but they show how rights holders value their catalogs when the alternative is a licence.

What does AI training data cost?

Most contracts stay private. These disclosed figures are the visible edge of a much larger market:

• $1 to $4 per minute: quality licensed video footage, with premiums for 4K, drone and exclusive content
• ~$40M per character: reported figure in Google's studio talks
• $60M to $70M a year: Reddit's text licensing run-rate
• $250M+ over five years: News Corp to OpenAI
• $16.3B by 2033: projected AI training data market across all modalities, about 22.6% CAGR (Grand View Research)

Treat these as directional. Figures marked reported are not contract-confirmed.

What this means for AI teams buying training data

1. Rights-clean beats raw volume. The categories with the highest proven spend are the ones where legal access is hardest. Volume you cannot defend is a liability, not an asset.
2. Provenance is becoming a requirement. Anthropic, OpenAI, Google and Microsoft now back one shared robots.txt AI-training opt-out. The New York Times and other papers have sued Brave Software over reselling scraped journalism to AI firms. In the EU, an Advocate General opinion found chatbot answers can infringe press publishers' rights. Where your data came from is getting more expensive to ignore.
3. Scarcity sets the price. If your model needs narrative, real-world or egocentric video, expect competition for supply and plan sourcing early.

What this means for video libraries and content owners

1. Genre matters more than hours. Scripted, exclusive or real-world footage sits at the top of the demand curve. Generic B-roll sits at the bottom, whatever the volume.
2. The market is pricing licensing and litigation at the same time. Rights holders who move first set the benchmarks everyone else negotiates against.
3. Readiness is the bottleneck. Buyers want rich metadata, accessible delivery and a verifiable chain of custody. Libraries that can provide all three get evaluated first.

Want to know where your library sits on this curve? Get a free Content Value Assessment.

Why Versos built the Demand Tracker

Versos is a data exchange connecting AI builders with premium video libraries. We give rights holders the infrastructure to index, prepare, license and deliver their content, with C2PA provenance and a full chain of custody built in. You can browse the libraries in the Versos Library Network.

We built the tracker because both sides of the market are making decisions with too little information. It applies the same discipline we use on video data to the demand side: public sources, scored consistently, kept current.

Frequently asked questions

What is AI training data demand?

It measures which kinds of data AI developers are actively seeking and paying to license. The Versos tracker scores it on capability heat, scarcity and proven spend.

Which type of AI training data is most in demand in 2026?

As of November 2026, scripted film and TV video ranks highest with a demand index of 95, followed by forums and community text (88) and real-world world-model video (85).

How much does AI training data cost?

Disclosed signals put quality licensed video at $1 to $4 per minute, with premiums for 4K, drone and exclusive footage. Major text deals run from tens to hundreds of millions of dollars. Most contracts are private, so treat public figures as directional.

How big is the AI training data market?

Grand View Research projects the AI training data market will reach $16.3B by 2033, growing about 22.6% a year across all modalities.

Where does the tracker's data come from?

Only from named public sources, including Variety, TechCrunch, Press Gazette, Nieman Lab, Rolling Stone and company announcements. No Versos customer or private data is used.

How often is the tracker updated?

It refreshes continuously as new public evidence lands. The latest refresh date is shown at the top of the tracker.

How can I license my video library for AI training?

Start with a Content Value Assessment to see how your library maps to current demand, then talk to Versos about preparing, licensing and delivering it with full provenance.

Explore the live AI Training Data Demand Tracker