Runway’s CEO thinks AI video is just the warm-up act for world models

AI-generated video has gone from a party trick to a legitimate creative tool almost overnight, and Runway has had a front-row seat the whole time. The New York-based company has raised close to $860 million at a $5.3 billion valuation, and its models are going toe-to-toe with the most well-funded labs in the world, including Google and OpenAI. That’s no small feat for a company that started as a research project in a Brooklyn apartment.

But here’s the thing: Runway’s CEO, Cristóbal Valenzuela, doesn’t think video is the endgame. He sees it as a prequel. In a recent interview, he laid out a vision that goes way beyond generating clips of astronauts on horseback or melting clocks. He’s talking about world models — systems that don’t just generate pixels but understand physics, causality, and how environments behave.

This is a shift that’s been brewing for a while. If you’ve been paying attention to the AI research papers coming out of places like DeepMind or Meta’s FAIR lab, you’ve seen the term “world model” pop up more and more. The idea is simple on the surface: instead of training a model to predict the next frame in a video, you train it to understand the underlying rules that govern that video. Why does a ball bounce? What happens when a glass tips over? How does light change when a cloud passes overhead?

Current AI video models, including Runway’s own Gen-3 Alpha, are essentially very sophisticated pattern matchers. They’ve seen enough videos of people walking that they can generate a plausible walk, but they don’t actually know what legs are or why gravity pulls things down. That’s fine for short clips and creative experimentation, but it falls apart when you need consistency, physical accuracy, or long-term coherence.

Valenzuela argues that the next logical step is to build models that learn those rules directly. Instead of training on text-to-video pairs, you train on video alone, letting the model infer the physics from observation — much like a child learns that dropped objects fall. It’s a different paradigm, and it’s one that Runway has been quietly investing in.

I find this direction more interesting than the race to make video models faster or higher resolution. Resolution is a solved problem at this point; we can already generate 4K clips that look photorealistic. The harder problem is making those clips obey the laws of physics without explicit programming. If Runway can crack that, they’re not just competing with OpenAI and Google on video generation — they’re building a foundation for simulation, robotics, and even game engines.

Of course, there’s a long road ahead. World models are computationally expensive, data-hungry, and still prone to hallucinating physics (I’ve seen plenty of demos where objects phase through each other or gravity suddenly reverses). But the ambition is real, and the funding is there. At $5.3 billion, Runway has the runway — pun intended — to take some big swings.

What I appreciate about Valenzuela’s take is that he’s not overselling. He acknowledges that AI video today is impressive but limited. He’s not claiming world models are around the corner. He’s saying they’re the natural next step, and Runway wants to be the one to take it. That’s a refreshingly honest position in an industry full of hype cycles.

So yeah, AI video is fun. It’s useful for storyboarding, prototyping, and making weird memes. But if Runway’s bet pays off, we might look back at this era the way we look back at flip phones — a necessary stepping stone, but laughably primitive compared to what came next.