Origin Lab Raises $8M to Turn Video Games Into AI Training Data

As artificial intelligence begins to step off the screen and interact with the physical world, a new breed of AI labs is scrambling for the right data to teach their systems how reality actually works. Origin Lab, a San Francisco-based startup announced an $8 million seed funding round on May 12, 2026, to unlock an unlikely treasure trove of training data: the video game industry.

Unlike the massive language models that learned to talk by scraping abundant text from the internet, the next frontier of AI involves "world models." These advanced systems are designed to operate physical robotics, model objects in physical space, and predict how things move and respond to forces. Development has stalled because researchers find it difficult to scrape clean, physical-world training data from the web. Co-CEO and co-founder Anne-Margot Rodde explained that the AI systems being built today desperately need to understand physical realities. She noted that this exact kind of data essentially lives in video games, as game studios have spent decades building rich, physics-based environments.

To fix this massive bottleneck, Origin Lab is building a unique marketplace that serves as a bridge between gaming and artificial intelligence. On one side, video game publishers can squeeze additional revenue out of the digital assets they have already created and released. On the other side, top-tier AI researchers, such as Yann LeCun’s AMI Labs and Fei-Fei Li’s World Labs, can buy high-quality, legally licensed data. 

In the middle sits Origin Lab, handling the complex technical work of converting 3D environments, physics simulations, non-player character behaviors, and camera telemetry into machine-learning-friendly formats. This translation process ranges from simple rendering runs to automating hours of high-fidelity walkthrough footage.

Heavy-Hitting Investors Target the Data Bottleneck

This compelling premise attracted heavy-hitting investors from across both the tech and gaming landscapes. The $8 million USD seed round was led by Lightspeed Ventures, with participation from SV Angel, Eniac, Seven Stars, and FPV. The startup, founded by Rodde alongside Antoine Gargot and Colin Carrier, also secured prominent angel investments from Twitch co-founder Kevin Lin and Cruise founder Kyle Vogt, as per TechCrunch. 

Faraz Fatemi, a partner at Lightspeed who led the investment, emphasized that while major AI labs are extremely well-capitalized businesses, data remains their primary bottleneck. Fatemi pointed out that the success of data vendors like Scale AI proves just how sharp the revenue scaling can be for startups serving as essential suppliers to these tech giants.

The demand for ethically sourced, licensed data has never been higher, especially following public backlash over unregulated web scraping. In December 2024, OpenAI faced a minor scandal when the first version of its Sora video-generation model appeared to regurgitate copyrighted video game content and popular streamers, presumably because it was trained on Twitch streams. 

While tech giants like Amazon have been open about their interest in using Twitch footage for AI training, Origin Lab will now provide a legally sound alternative that eliminates copyright risks and ensures fair attribution. 

The sheer scale of this highly competitive market for spatial training data was highlighted in October 2025, when OpenAI reportedly offered $500M for the data rights to gaming platform Medal. Medal founder Pim de Witte discovered the immense value of this asset class when he began selling his platform's data to AI labs, noting that Medal receives two billion gaming video uploads per year across tens of thousands of titles. 

With its fresh capital injection and growing infrastructure, Origin Lab is perfectly positioned to turn the broader gaming industry's untapped archives into the next great gold mine for AI development.