Runway Extends Video Models to Robots

This move turns Runway from a tool for making synthetic video into a supplier of the training inputs robotics teams cannot easily collect themselves. The same core strength that helped it automate editing, learning how motion changes frame by frame, maps cleanly into robotics, where developers need models that can predict what a robot should do next and simulators that let them test policies before touching real hardware.1 2 3 4 5

• Runway is attacking the two hardest robotics data problems at once. Praxis-1 is a pretrained policy model that developers can fine tune for a specific robot or task, while GWM Robotics simulates scenes and actions to generate synthetic training data and evaluate policies without running physical machines.2 3 4
• This is consistent with how Runway has always built. It started by using machine learning to remove repetitive frame by frame work in video, then moved up the stack into production workflows. Robotics is a new customer, but the technical pattern is similar, model driven automation for motion heavy work.5 6
• The strategic payoff is distribution beyond media. Runway already has roughly $200M ARR as of August 2026, and early robotics partner Standard Bots reached $24M annualized revenue in 2025. If robotics teams adopt Runway models as shared infrastructure, it gains a second growth lane next to filmmaking and advertising.1 4 7

Going forward, the winner in physical AI may be the company that owns synthetic data and evaluation before it owns the robot. Runway is positioning for exactly that layer. If Praxis-1 and GWM Robotics become standard starting points for developers, Runway can become embedded in robotics workflows the way CUDA became embedded in AI training.2 3 4