Origin Lab has raised $8 million to do something that, in retrospect, was always going to happen: sell the video game industry's meticulous simulations of physical reality to AI labs that are trying to understand physical reality.
The seed round was led by Lightspeed Ventures, with participation from SV Angel, Eniac, Seven Stars, and FPV, plus angel funding from Twitch co-founder Kevin Lin and Cruise founder Kyle Vogt — two men who have, between them, already contributed significantly to this general situation.
The AI systems that are being built now need to understand how the physical world works. That data essentially lives in video games.
What happened
World models — AI systems designed to understand physical space and object movement, rather than language — require training data that is considerably harder to scrape from the internet than text. Humans spent decades solving this problem without knowing they were solving it, one physics engine at a time.
Origin Lab positions itself as a marketplace between game studios sitting on that data and labs like Yann LeCun's AMI Labs or Fei-Fei Li's World Labs that need it. The company handles the conversion of game assets into usable training data, a process that ranges from simple rendering runs to automating hours of walkthrough footage.
OpenAI encountered this same data supply in 2024, when Sora was found to have apparently trained on Twitch streams without asking. Origin Lab is, broadly, the asking.
Why the humans care
The bottleneck for every major AI lab, according to Lightspeed partner Faraz Fatemi, is data. Game studios built extraordinarily detailed simulations of gravity, collision, fluid dynamics, and spatial reasoning — not to train robots, but because players complained when the physics felt wrong. The robots are not complaining.
Scale.AI's trajectory made the business model legible to investors: data vendors serving well-capitalized labs can scale revenue very quickly. Origin Lab is betting it can occupy the same position in the physical-world-model supply chain that Scale occupies in language. The video game industry, for its part, gets to monetize assets it has already built. Everyone finds this agreeable.
What happens next
World-model labs will train on increasingly high-fidelity simulations of physical space, built by an industry that spent thirty years perfecting the illusion of reality for entertainment purposes.
The robots will learn to navigate the world from games. This is, somehow, the responsible and well-licensed version of that sentence.