As AI starts interacting with the physical world, new labs are working to create world models for operating robotics or modeling objects in physical spaces. Unlike large language models, there isn’t a straightforward source of data for these models, leading many labs to scramble for training sets.
One startup is turning to an unexpected data source: the video game industry.
This is the concept behind Origin Lab, a startup that announced an $8 million seed funding round led by Lightspeed Ventures, with participation from SV Angel, Eniac, Seven Stars, and FPV, along with angel funding from Twitch co-founder Kevin Lin and Cruise founder Kyle Vogt.
“The AI systems being built need to understand the physical world and how things move,” said co-CEO and co-founder Anne-Margot Rodde. “That data essentially lives in video games.”
Origin Lab will act as a marketplace where world-model-focused labs, such as Yann LeCun’s AMI Labs or Fei-Fei Li’s World Labs, can acquire high-quality licensed data. Video game companies can generate additional revenue from their digital assets. Origin Lab will convert these video game assets into suitable training data, which could range from simple rendering runs to complex automated walkthrough footage.
“There was incredibly valuable data in the video game industry, but no infrastructure to connect AI labs and the gaming sector,” Rodde explained. “So we built that bridge.”
Labs have been interested in video game footage as a data source, but issues with licensing and quality have been barriers. In December 2024, OpenAI faced controversy when its Sora video-generation model appeared to use popular video game footage, likely trained on Twitch streams. Amazon has also shown interest in utilizing Twitch footage for model training.
Origin’s successful fundraising reflects a growing market for training data and startups serving as key suppliers to AI labs. Faraz Fatemi, a partner at Lightspeed who led the investment, highlighted the growing opportunity for data vendors supporting major labs.
“We’ve seen sharp revenue scaling for data vendors working with major labs,” Fatemi said. “These well-capitalized businesses are all bottlenecked by data.”