As artificial intelligence moves beyond text and images into the physical world, a new bottleneck has emerged: high-quality training data for world models. These models, designed to simulate and understand physical environments for robotics, autonomous systems, and spatial computing, require vast amounts of data that capture real-world physics, object interactions, and movement. Unlike large language models, which can be trained on text scraped from the internet, there is no readily available repository for physical world data. One startup, Origin Lab, believes the answer lies inside the video game industry.
A marketplace for synthetic physical data
Origin Lab has announced an $8 million seed funding round led by Lightspeed Venture Partners, with participation from SV Angel, Eniac, Seven Stars, and FPV. Angel investors include Twitch co-founder Kevin Lin and Cruise founder Kyle Vogt. The company plans to build a marketplace where AI labs focused on world models can purchase licensed, high-quality data extracted from video games. On the other side, video game companies can generate new revenue streams from existing digital assets, such as 3D environments, character models, and physics engines.
Also read: What the jury will actually decide in Elon Musk vs. Sam Altman
“The AI systems that are being built now need to understand how the physical world works and how things move,” said Anne-Margot Rodde, co-CEO and co-founder of Origin Lab, in an interview with TechCrunch. “That data essentially lives in video games.”
Origin Lab will act as a middle layer, converting raw video game assets into formats suitable for training world models. This conversion can range from simple rendering runs to complex automation of hours of walkthrough footage, capturing object interactions, lighting, physics, and spatial relationships.
Also read: SpaceXAI Bleeds Talent: Over 50 Researchers Depart Since Merger, Raising Doubts About AI Ambitions
Why video games are a goldmine for AI training
Video games are built on sophisticated physics engines that simulate gravity, collision, friction, and object behavior. They also contain meticulously designed 3D environments and assets that mirror real-world geometry and lighting. For AI labs building world models — such as Yann LeCun’s AMI Labs or Fei-Fei Li’s World Labs — this data is invaluable. It provides labeled, structured, and controllable examples of how objects move and interact in space.
However, licensing and data quality issues have historically hindered access. In December 2024, OpenAI faced scrutiny when its Sora video-generation model appeared to reproduce footage from video games and Twitch streams, suggesting training on unlicensed data. Amazon has also publicly expressed interest in using Twitch footage for model training. Origin Lab aims to solve this by providing a legitimate, licensed pipeline.
Investor confidence in data infrastructure
Faraz Fatemi, a partner at Lightspeed who led the investment, highlighted the precedent set by companies like Scale AI, which has grown into a multi-billion-dollar business by supplying training data to major AI labs. “We’ve seen how sharp the revenue scaling can be for data vendors that are serving the major labs,” Fatemi told TechCrunch. “These are very well-capitalized businesses, and the bottleneck for all of them is data.”
The investment signals growing market confidence in specialized data infrastructure startups that serve the AI industry’s expanding need for diverse, high-quality training datasets.
Why this matters
The emergence of Origin Lab reflects a broader shift in the AI sector. As foundational models for language and vision become commoditized, the competitive advantage is moving toward data — specifically, data that is difficult to scrape or generate. Physical world data, in particular, remains scarce and expensive to collect. By tapping into the existing digital assets of the video game industry, Origin Lab could accelerate the development of world models, which are critical for applications in robotics, autonomous vehicles, augmented reality, and industrial automation.
For video game companies, the opportunity is equally significant. Many studios sit on vast libraries of 3D assets and game environments that are no longer generating revenue. Origin Lab’s marketplace offers a way to monetize these assets without disrupting core game development or publishing operations.
Conclusion
Origin Lab’s $8 million seed round is a bet that the video game industry holds the key to solving one of AI’s most pressing data challenges. By building a legitimate marketplace and data conversion pipeline, the startup aims to unlock a new category of training data for world models. If successful, it could reshape how AI labs access physical-world data and provide a new revenue stream for game developers. The coming months will reveal whether this bridge between two industries can scale to meet the demands of the world’s most ambitious AI projects.
FAQs
Q1: What are world models in AI?
World models are AI systems designed to understand and simulate the physical world. They model how objects move, interact, and behave in space, and are used for training robotics, autonomous vehicles, and other physical AI applications.
Q2: Why is video game data valuable for training AI world models?
Video games contain sophisticated physics engines, 3D environments, and labeled interactions that simulate real-world physics. This data is structured, controllable, and abundant — ideal for training models that need to understand spatial relationships and object dynamics.
Q3: How does Origin Lab plan to make money?
Origin Lab will operate as a marketplace, taking a cut of transactions between AI labs purchasing data and video game companies selling licensed assets. The company also provides data conversion services to prepare game assets for use as training data.
Be the first to comment