Runway’s CEO thinks AI video is just the appetizer — world models are the main course

Runway has been on a hell of a run. 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 not nothing for a company that started as a creative tool for artists.

But CEO Cristóbal Valenzuela isn’t satisfied with just making better AI video. In a recent interview, he argued that video generation is really just a prequel to something much bigger: world models. The idea is that these models don’t just generate pixels — they learn the underlying rules of physics, geometry, and causality. So instead of just making a clip of a cat walking, a world model could understand that the cat has mass, that it casts shadows, that it can’t walk through walls.

This is the kind of thinking that separates Runway from the herd. Most AI video companies are obsessed with resolution, frame rate, and prompt adherence. Runway seems to be asking: what if we built a model that actually understands the world it’s simulating? I’ve been watching this space for a while, and I have to say, it’s a more ambitious bet than just chasing better Gen-3 benchmarks.

Valenzuela is careful not to overpromise. He admits world models are still in early stages, and that current AI video is more about interpolation than true understanding. But the direction is clear: Runway wants to build a model that can predict what happens next in a scene, not just fill in the blanks between frames. That’s a fundamentally different approach from what most labs are doing.

The practical implications are huge. If you can simulate physics accurately enough, you could use world models for robotics training, autonomous driving, architectural visualization, or even scientific discovery. Imagine training a robot in a simulated world that actually follows real-world physics, without needing millions of dollars in hardware. That’s the kind of thing that makes VCs open their checkbooks.

But I’m also a little skeptical. We’ve heard this kind of talk before. Every few years, someone claims their AI has achieved “understanding” rather than just pattern matching. The hype cycle around world models is real, and it’s easy to get carried away. Runway’s own demos are impressive, but they still break when you push them too far — objects disappear, physics glitch, lighting fails. Calling that a “world model” feels generous.

Still, Runway has a track record of shipping. They’ve been iterating fast, and their products are actually used by working creatives, not just tech demos. That’s more than I can say for some of the bigger labs. If anyone can push video generation toward genuine world simulation, it might be a company that started by helping editors cut trailers.

The real question is whether the market will reward this ambition before the money runs out. $860 million is a lot, but it’s not infinite. Building a world model from scratch is probably a decade-long project, and investors aren’t known for their patience. Runway needs to keep selling video tools to stay alive while they chase the bigger vision.

I think Valenzuela is right about the direction, even if the timeline is fuzzy. AI video is a stepping stone, not a destination. The companies that treat it as an end in itself will get left behind when the real shift happens. Runway is betting that they can be the ones who make that shift. It’s a high-risk, high-reward move, and I’m genuinely curious to see how it plays out.