Origin Lab has raised an $8 million seed round led by Lightspeed Ventures to build a marketplace that licenses video game data to AI labs training world models. SV Angel, Eniac, Seven Stars, and FPV joined the round, with angel checks from Twitch co-founder Kevin Lin and Cruise founder Kyle Vogt. The pitch is straightforward: world models need physics-grounded video data, and game studios are sitting on petabytes of it with no clean way to sell it.
World models — the systems behind robotics planners and video generators like Sora — need training data that captures how objects move, collide, and persist over time. Text scraped from the web does not teach that. Game engines, by contrast, produce continuous, labeled, physically consistent 3D footage at scale, and studios already own the rights.
"The AI systems that are being built now need to understand how the physical world works and how things move," co-CEO and co-founder Anne-Margot Rodde told TechCrunch. "That data essentially lives in video games."
Key facts
- 01Origin Lab closed an $8 million seed round led by Lightspeed Ventures, with SV Angel, Eniac, Seven Stars, and FPV participating.
- 02Angel checks came from Twitch co-founder Kevin Lin and Cruise founder Kyle Vogt.
- 03The startup is a marketplace connecting video game studios to world-model labs like Yann LeCun's AMI Labs and Fei-Fei Li's World Labs.
- 04OpenAI's first Sora release in December 2024 appeared to regurgitate video game and Twitch footage, exposing the licensing gap Origin Lab targets.
- 05Lightspeed partner Faraz Fatemi cites Scale AI's growth as evidence that data vendors to major labs can scale revenue sharply.
Origin Lab's customers on the buy side include Yann LeCun's AMI Labs and Fei-Fei Li's World Labs — two of the highest-profile world-model efforts to launch in the past year. On the sell side are game publishers and studios who can monetize digital assets they have already built. Origin Lab handles the conversion layer in the middle, which can range from a basic rendering pass to automating hours of in-engine walkthrough footage tailored to a lab's training spec.
“We've seen how sharp the revenue scaling can be for data vendors that are serving the major labs. These are very well-capitalized businesses, and the bottleneck for all of them is data.”— Jaeden Schafer
"It became clear that the video game industry was sitting on some incredibly valuable data, but there was no real way or infrastructure to basically connect AI labs and the video game industry," Rodde said. "So essentially, we built that bridge."
The licensing gap is not theoretical. In December 2024, the first version of OpenAI's Sora drew scrutiny when generated clips appeared to reproduce footage of popular video games and Twitch streamers, suggesting the model had ingested stream data without clear permission. Amazon, which owns Twitch, has been open about its interest in using stream footage for model training. Origin Lab's bet is that labs would rather pay for clean, licensed, format-ready data than litigate scraped corpora.
Lightspeed partner Faraz Fatemi, who led the deal, frames Origin Lab as part of a broader category of suppliers to frontier labs. "We've seen how sharp the revenue scaling can be for data vendors that are serving the major labs," Fatemi said. "These are very well-capitalized businesses, and the bottleneck for all of them is data." The reference point is Scale AI, which built a multibillion-dollar business labeling data for OpenAI, Anthropic, and Meta.
The competitive question is whether game publishers will route through a marketplace at all, or strike direct deals with labs. Large studios with valuable engine data — open-world franchises in particular — have the leverage to negotiate one-on-one and may not need a broker. Origin Lab's edge has to be in the conversion layer and in aggregating long-tail studios that individually cannot afford a sales motion to AI labs.
There is also an unsettled question of what game data is actually worth per hour to a world-model lab, and whether the supply will outrun the demand once a few large licensing deals close. World-model training is still early, and the labs buying today — AMI Labs, World Labs, and a handful of robotics outfits — are a small customer base compared with the dozens of language-model shops Scale AI serves.
For Origin Lab, an $8 million seed is enough to prove the marketplace mechanics with a few anchor studios and a few anchor labs. If world models become the next training-compute arms race — and the early funding signals from LeCun and Li suggest the major investors believe they will — the data-broker layer is going to matter as much for embodied AI as Scale AI has for chatbots. The opening is that no one else is building this bridge yet, and the labs paying for the data would rather not build it themselves.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
