Conference Presentation, Fireside Chat, Interview
Fireside Chat with Anastasis Germanidis, Co-Founder & Co-CEO of Runway | RAISE Summit 2026
- The company predicts generative video models will lead "generative for world models" by scaling compute and data to become powerful simulators that learn 3D consistency and physics directly from 2D visual data without explicit 3D representations.
- Real-time video models are expected to enable robotics and gaming applications previously impossible with traditional methods, offering improved trajectory accuracy for dropped objects and more effective simulation of manipulation tasks like folding clothes.
- The company maintains a thesis that "next frame prediction" is the most effective approach for video models and anticipates these models will predict robot actions more accurately than traditional methods like VLA's, requiring only tens or hundreds of hours of robotics data compared to hundreds of thousands or millions.
- Enterprise adoption is forecast to change significantly this year, with video models moving beyond pre-production to generate final frames and enable real-time interactive experiences such as live-responding avatars.
- Video models will allow customers to correlate simulation testing well with real-world performance to accelerate model development, while the company plans to license these models directly to robotics customers to serve as powerful policy models.
- Operational improvements are expected as real-time models become cheaper to run and further efficiency gains are made in training and inference, though the company advises startups to stay small and grow intentionally to maintain long-term focus in a dynamic field.
- Strategic expansion includes opening a new office in Paris to invest in research talent and partner with customers in the European physical AI ecosystem, citing local AI growth, government investment, and customer presence as key drivers.