
A comprehensive intelligence brief for AAA FPS studio executives on companies actively seeking gameplay footage to train large frontier world models — and what they are willing to pay.
The AI training data market is undergoing a structural transformation. As large language models approach saturation on text data, the next frontier — world models — requires something fundamentally different: rich, interactive, spatially complex video data.
First-person shooter footage is uniquely valuable in this landscape. FPS games provide continuous first-person perspective data — the same viewpoint used by autonomous vehicles, drones, and robotic agents. The complex physics interactions, spatial navigation, and action-consequence relationships encoded in FPS gameplay are precisely what frontier world models need to develop genuine physical intuition.
OpenAI's reported $500M offer to acquire Medal.tv — a platform with 2 billion FPS and action game clips per year — is the clearest signal of how much AI labs value high-quality gameplay footage. Your AAA FPS studio is sitting on a comparable or superior asset.
Source: Precedence Research. E = Estimate, P = Projected. Values in USD billions.
OpenAI offered $500M to acquire Medal.tv for its 2B gameplay clips/year dataset.
Direct dataset acquisition, studio partnerships
Negotiated bulk licensing
FPS footage provides ideal first-person spatial-temporal data. CEO Pim de Witte stated games are 'the equivalent of verifiable domain for spatial temporal reasoning.'
Offered $500M to acquire Medal.tv solely for its gameplay footage repository.
Direct video licensing, major acquisitions
$1–$4 per minute, exclusive deals
Sora and successor video models require diverse, high-action footage. FPS games provide complex motion, physics interactions, and environmental diversity unavailable in typical video datasets.
Genie 3 powers Waymo's World Model for autonomous driving simulation — trained on diverse video including game footage.
Exclusive licensing, developer partnerships
Negotiated enterprise licensing
Genie 3 was trained on 'an extremely large and diverse set of videos.' Waymo's use of Genie 3 for autonomous driving simulation demonstrates direct value of game-like first-person spatial footage.
Raised $20B and is actively hiring Nvidia world model experts. Musk confirmed AI-generated game development.
Data licensing, studio partnerships
Negotiated, premium for exclusive
xAI is explicitly building world models 'initially for gaming' — generating immersive 3D environments. Elon Musk confirmed plans to release 'a great AI-generated game' and is hiring game-focused AI researchers.
Published 'Muse' in Nature — first World and Human Action Model trained on 7+ years of continuous gameplay from a single title.
Research collaboration, internal licensing
Revenue sharing, research partnership
Microsoft's Muse (WHAM) model was trained on Bleeding Edge gameplay footage via a Ninja Theory partnership. They are actively seeking to expand to more game genres, including FPS.
Hunyuan-GameCraft trained on 1M+ gameplay recordings across 100+ AAA games. Now open-sourced.
Internal utilization, potential external licensing
Negotiated enterprise deals
Hunyuan-GameCraft was trained on 1,600+ hours from 100+ AAA titles. The research paper notes the dataset includes 'first and third person gameplay with complex agent behaviors and physics.'
Founded by Fei-Fei Li (Stanford AI pioneer). Marble API launched Jan 2026. Partnered with NVIDIA Isaac Sim.
Data partnerships, studio collaborations
Negotiated partnership
World Labs' Marble product generates 3D worlds from video input. FPS footage provides rich first-person spatial data for training their spatial intelligence models.
Backed by NVIDIA. First programmable world model for real-time interactive content. 10,000+ waitlist users.
Data licensing, co-development
Revenue sharing, licensing
Moonlake's Generative Game Engine is specifically designed for real-time interactive content. FPS footage would directly train their world model for first-person game generation.
Current market rates for video footage licensing range dramatically based on exclusivity, volume, and strategic value. As a AAA FPS studio, you are positioned at the premium end of this spectrum.
General creator content
Premium gameplay content
Bulk exclusive licensing
Strategic acquisitions
Note: xAI $20B represents total raised capital for world model development, not a single data deal.
If direct outreach to AI labs feels premature, data brokers and marketplaces offer a lower-friction path to monetization. These platforms handle rights clearance, ingestion, and buyer matching — allowing studios to generate revenue from existing footage libraries without dedicated business development resources.
Rights-cleared marketplace connecting studios directly with AI developers
Studios wanting passive income from existing footage libraries
Supports LLMs, world models, and foundational architectures. FedRAMP authorized.
AI-ready video archive platform with natural-language search
Studios with large unstructured footage archives
Makes your entire library discoverable by AI buyers without manual tagging.
Data labeling and AI training data provision for major AI labs
Studios wanting to add annotation value to raw footage
Meta invested $14B. Direct pipeline to OpenAI, Meta, and other frontier labs.
Catalog your existing gameplay recordings, including raw capture data, player telemetry, and controller input logs. FPS footage with synchronized controller actions (like Microsoft's Muse training data) commands premium pricing. Quantify hours, resolution, and game diversity.
Ensure you have clear, transferable rights to the footage — including player consent (via EULA), third-party asset clearances, and music licensing. Microsoft's Ninja Theory partnership succeeded because they built consent into the EULA from the start. Retroactive clearance is costly.
These two companies represent the highest urgency and most direct fit for FPS footage. General Intuition explicitly seeks first-person game data; xAI is building a gaming world model with $20B in capital. Both are in active data acquisition mode as of Q1 2026.
Exclusive deals command 10–50x premiums over non-exclusive licensing. However, exclusivity forecloses future deals. Consider time-limited exclusivity (12–24 months) or field-of-use restrictions (e.g., exclusive for autonomous driving, non-exclusive for gaming AI).
Veritone Data Marketplace (launched March 2026) and Versos offer low-friction paths to monetize existing footage libraries. These platforms handle rights clearance and buyer matching, allowing your team to focus on game development while generating data licensing revenue.
AI data licensing is a rapidly evolving legal landscape. Engage IP counsel familiar with AI training data agreements before entering any negotiations. Key issues include: output licensing restrictions, model attribution, data deletion rights, and indemnification clauses.
The legal landscape for AI training data is rapidly evolving. Tencent's Hunyuan-GameCraft — trained on 100+ AAA titles — raises immediate questions about data acquisition that the published research doesn't address. OpenAI's Sora has faced scrutiny for training on copyrighted game content.
This legal uncertainty is a competitive advantage for studios willing to offer properly licensed footage. AI labs are actively seeking "rights-cleared" data to avoid litigation risk, and are willing to pay significant premiums for it.
Restrict AI companies from generating content that competes directly with your game IP.
Negotiate the right to require deletion of training data if the partnership ends.
Ensure your EULA covers commercial use of gameplay recordings for AI training.
Consider ongoing royalties tied to model performance rather than one-time licensing fees.