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

AI-generated video has gone from party trick to 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. That puts them in the same weight class as the most well-funded labs on the planet — Google, OpenAI, you name it. And their models are actually going toe-to-toe with those giants, which is no small feat for a startup that started as a research project.

But here’s the thing that caught my attention: CEO Cristóbal Valenzuela doesn’t seem satisfied with just making better video generators. He’s been talking about “world models” — systems that don’t just predict the next pixel, but simulate how the physical world actually works.

What’s a world model, really?

The term gets thrown around a lot, so let me be specific. A world model isn’t just a bigger, faster video generator. It’s an AI that learns the rules of physics, causality, and spatial relationships from visual data. Think of it as an internal simulation engine: you show it a ball rolling toward a wall, and it can predict the bounce without ever being explicitly programmed with Newton’s laws.

Runway’s approach builds on their video generation work, but the goal is fundamentally different. Video generation cares about producing convincing frames. A world model cares about producing a coherent understanding of the scene itself — what exists, how things interact, what happens next.

Valenzuela put it bluntly in a recent interview: “Video generation is the prequel.” That’s a characteristically confident statement, but I think he has a point. If you can generate video that respects physics and object permanence, you’re already halfway to a world model.

The money and the moat

$860 million is a lot of runway — pun intended — but it’s not infinite. Runway is burning cash on compute, talent, and research. They need to show that world models are more than a research curiosity before investors get impatient.

What gives me some confidence is that they have a real product today. Creators are using Runway’s tools for music videos, commercials, even film projects. That’s actual revenue, not just demo reels. Compare that to some of the more theoretical world model efforts coming out of academia, and Runway looks like they have a much clearer path to practical application.

That said, the competition is fierce. Google’s DeepMind has been working on world models for years. OpenAI has their own video generation efforts. And there are smaller labs like Sora (not the OpenAI one) that are pushing in similar directions. Runway’s edge might be their focus on creative tools — they understand the user, not just the algorithm.

Where I get skeptical

I’ve been around long enough to see plenty of “next big things” in AI that never materialized. World models have been a research topic since at least the 1990s, and we’re still not at the point where a model can reliably simulate a simple physics experiment, let alone an open world.

Runway’s demos look impressive, but demos always look impressive. The real test is whether these models can generalize to situations they haven’t seen before. Can a world model trained on video of falling objects handle a scenario where the object is made of rubber vs. steel? Can it handle friction? Air resistance? Those are the kinds of questions that separate a toy from a tool.

I also worry about the compute requirements. World models are fundamentally more expensive than video generators because they need to maintain internal state and run simulations. If the cost per inference is too high, the applications will be limited to well-funded studios and research labs — not the independent creators Runway claims to serve.

What this means for creators

If Runway pulls this off, the implications are huge. Imagine a video editing tool where you can place a virtual object in a scene and have it behave realistically — no keyframing, no physics sim, just the AI handling it. Or a game where the world reacts to player actions without pre-scripted animations. That’s the promise.

But let’s be real: we’re probably years away from that. The immediate future is more incremental — better video generation, more control, faster rendering. World models will arrive in pieces, not all at once.

Runway is betting that the pieces fit together. I’m not convinced yet, but I’m watching closely. The fact that they’re willing to talk about the long-term vision while shipping real products is more than most AI companies can say.

And honestly? That’s more than I expected from a startup that started as a research project in a New York apartment.