Runway wants to turn video generation into world models and surpass Google in the AI race

Runway, known to the market for its video generation tools for creators and studios, is attempting to reach a much larger goal: build world models and compete with Google not just in video, but in the next layer of AI. For the company, this is not a side experiment, but a new core bet.

Betting on Video

While most of the industry bets on language models, Runway's co-founders believe that text alone is insufficient. By their logic, large language models learn from how people describe the world, not from the world itself. Video, on the other hand, contains observations of movement, physics, causality, and environmental change over time.

This is precisely why Runway views video generation not as an end product for editing, but as an intermediate stage on the path to systems that can predict environmental behavior. Runway co-founder Anastasis Germanidis frames this even more harshly: if AI is limited to human descriptions, it hits a ceiling of already accumulated knowledge. Runway wants to teach models on more direct observational data and through this approach move closer to systems that not only draw convincing frames but truly understand how the world is structured.

For the company, a world model is a system that can simulate an environment and forecast what will happen next in a particular situation.

From Film to Platform

Runway was founded in 2018 by three co-founders who met at NYU and initially dreamed of cinema. The company's first question was simple: can AI make every person capable of making a film? But after the release of early video models, the focus shifted.

Now the goal is broader: to make video generation the foundation for new computational systems that will be useful not only in cinema, but also in robotics, science, and applied simulations. Over recent months, Runway has already translated this idea from presentations into products. In December 2025, the company presented GWM-1—its first family of world models—while continuing to develop its video model Gen-4.

5. Against this backdrop, it is clear that this is not a distant research plan: the company already has commercial business, major clients, and organizational scale that allows it to fund a riskier bet.

• company valuation — $5.3 billion
• plus $40 million in annual recurring revenue added in Q2 2026
• 155 employees in offices from New York to Tokyo
• partnerships with Lionsgate and AMC Networks
• a separate division for robotics launched in 2025

Runway is also important because it is not a classic "laboratory" player from Silicon Valley. The company has no origins in the manner of former Google or Stanford alumni, and its co-founders directly state that they perceive their position as an outsider as an advantage. By their account, the early necessity to earn money forced the team to convert research into products faster, and the absence of Silicon Valley standardization allowed them to find their own path without unnecessary ready-made formulas.

Money and Competitors

Runway's main problem is not the idea, but the scale of the race. To move from strong video generation to full-fledged world models requires enormous computational resources, a long investment horizon, and access to infrastructure on par with leading laboratories. The company already collaborates with Nvidia and CoreWeave and in February 2026 raised another $315 million.

But this is still insufficient against players who can afford nearly endless cycles of experimentation. The most obvious threat is Google. It already has powerful video generation and its own world models direction, and with that—infrastructure, research capabilities, and capital reserves that are unreachable for a startup.

Runway also has to watch Luma, World Labs, and other teams, while proving the most important thesis: that the path from video to general understanding of the world is even possible. The bet is bold, but not yet proven.

"Rules are just rules that someone invented."

This thought from Runway co-founder Cristóbal Valenzuela well explains the company's style. Runway is not trying to play by the canon of big AI labs and bets on speed, product discipline, and an unusual starting point—creativity instead of text. The question now is whether this will be enough when the dispute moves from beautiful demos to competition for computing, data, and sustainable advantage. This is what they consider their main cultural advantage.

What This Means

If Runway is right, video generation will stop being a niche tool for marketing and film and will become a stepping stone to a more important market—systems that can model the physical world. For the market, this is an important signal: the next major battle in AI may not be about the best chat answers, but about machines' ability to predict and reproduce reality. This is already a different market and a different logic of leadership.