Jul 22, 2026, 1:34 PM ETConference Presentation, Fireside Chat, Interview

Fireside Chat with Anastasis Germanidis, Co-Founder & Co-CEO of Runway | RAISE Summit 2026

Runway, founded in 2019, is shifting from being a leader in generative video AI to a leader in generative "world models" that simulate the physical world.
The company's thesis posits that scaling compute and data on video models enables them to learn 3D consistency and physics purely through pixel prediction without explicit 3D architectural injections.
A "world model" is defined as a system that builds an environmental representation allowing for interactive action testing, contrasting with traditional video models that are non-interactive and prompt-based.
Runway released its first general world model (GWM1) early in the year, built upon their existing video model architectures.
The company differentiates itself by adhering to the principle that simple methods, specifically next-frame prediction, often outperform complex approaches involving explicit 3D representations or non-pixel space logic (like JEPA).
Runway asserts that physics and complex object interactions (e.g., folding clothes, pick-and-place) are learned emergently through scale rather than hardcoded rules, enabling more effective simulation of manipulation tasks than traditional simulation methods.

Runway's video models can be adapted to predict robot actions, utilizing vast pre-trained datasets of human video to reduce the need for specific robotics training data from hundreds of thousands of hours to tens or hundreds of hours.
The company has prioritized real-time inference, noting that moving generation time from minutes to 100 milliseconds drastically improves user iteration speeds and reduces operational costs.
Runway achieved superior performance on its Gen 4.5 video model using approximately 100 times less compute and 10 times fewer personnel than large labs like Google, attributing this to efficient post-training techniques and high-quality data curation.
Evaluation metrics show consistent improvement in physics simulation accuracy (e.g., object trajectories) across every generation of Runway's models as they scale.
The technology allows for the correlation of performance in a simulated world model directly with real-world robotic performance, solving a key bottleneck in scaling robotics data collection.

Runway reported its best quarter ever in Q2, expanding its headcount to nearly 200 employees across multiple global offices.
Enterprise adoption has accelerated, with video models moving beyond pre-production visualization to being integrated into final rendered film frames and advertisements.
New real-time applications include interactive avatars that allow users to live, on-the-fly interact with generated characters, enabling experiences that previously required years of game development.
Runway is seeing primary clusters of interest in robotics, gaming, and interactive storytelling.
The company now licenses its video models directly to robotics customers, allowing them to use the models as policy models tailored to their specific robotic embodiments.
Runway recently opened a new office in Paris to deepen its European presence, citing strong research talent in world models and robotics, supportive government infrastructure investments, and a high concentration of prospective customers.
The Paris team will focus primarily on research in world models and robotics while fostering closer partnerships with physical AI customers in the region.

Runway founder Anastasis advises startups in the physical AI space to "stay small for as long as possible" to maintain agility in a rapidly changing field.
The company emphasizes intentional growth and discipline over rapid expansion, despite the current influx of AI investment capital.
Runway values speed and the ability to pivot quickly, believing that being a small, focused lab allows for more efficient training and deployment strategies compared to larger competitors.