Runway, the New York-based AI video startup, is no longer content with making tools for filmmakers. According to TechCrunch AI, the company has pivoted its ambitions toward building world models, AI systems that learn how reality works from observation rather than text. That puts Runway on a collision course with Google, and it’s a bet that could reshape what “intelligence” even means in this industry.
What stands out here is the philosophical split. For years, the dominant assumption has been that intelligence lives in language. ChatGPT, Claude, and Gemini are all monuments to that thesis. Runway co-founder and co-CEO Anastasis Germanidis told TechCrunch AI he thinks that’s a ceiling, not a foundation. “We’re basically bound by our own understanding of reality,” he said. “Language models are trained on the entire internet, on message boards and social media, on textbooks, distilling the existing human knowledge. But to get beyond that, we need to leverage less biased data.”
From indie filmmaking tool to $5.3B contender
Runway’s origin story doesn’t fit the usual Silicon Valley template. Two Chilean founders and one Greek founder met at NYU’s Tisch School of the Arts in 2016. They launched the company in 2018 with a modest pitch: use AI to make everyone a filmmaker.
The traction has been real:
- Valuation now sits at $5.3 billion
- Added $40 million in annual recurring revenue in Q2 2026
- Tools used in films including “Everything Everywhere All At Once”
- Signed deals with Lionsgate and AMC Networks
- 155 employees across New York, London, San Francisco, Seattle, Tel Aviv, and Tokyo
The flagship Gen-4.5 model already powers production workflows for studios and ad agencies. That’s a healthy business on its own. But Germanidis and his co-founders see video generation as the on-ramp to something much bigger.
Why world models matter
A world model is an AI system that simulates environments well enough to predict how they’ll behave. Think of it as a physics engine that learned reality by watching it, not by reading about it. Runway launched its first one in December and plans to ship another this year.
The near-term applications are interactive entertainment, gaming, and robotics training. The long-term pitch is much bolder. Germanidis frames world models as scientific infrastructure, essentially a digital twin of the universe you could run experiments on faster than any lab could. “If we can build a better scientist than human scientists, we can accelerate progress in how we understand the universe and how we solve problems,” he told TechCrunch AI.
Runway has already spun up a robotics unit that’s seeing real-world testing. Drug discovery and climate modeling are on the roadmap.
The competitive squeeze
Runway isn’t alone on this path. Luma and World Labs are chasing the same vision. Google’s Genie world model points in exactly the same direction, and Google has the cash, the compute, and the research talent to dump on the problem.
That’s the risk Runway is staring at. If video-to-world-models really is the next frontier, the company has a head start and genuine momentum. If it loses the race, it gets outpaced by competitors with deeper pockets and ends up as a premium tool for video editors rather than the platform that cracked AGI from a new angle.
The broader implication for AI practitioners is worth sitting with. The industry has spent three years optimizing language models. A serious cohort of researchers now thinks that approach has a ceiling, and the next leap comes from training on raw observational data instead of human descriptions of it. Whether that’s right or wrong, the bet is being placed with real money and real product roadmaps.
Full story at TechCrunch AI.