Runway made its name by providing AI tools to filmmakers. Now the company wants to prove that video generation can be a much bigger way to simulate the world.
The next activity of the runway is actually not to create more beautiful clips. A New York startup is looking to use its creative AI business as the basis for its World Model, a system that can learn how its environment behaves and predict what will happen next. This puts the 155-employee company in the same conversation as Google DeepMind, OpenAI, and other labs with much deeper computing budgets.
Timing is critical. A May 15 profile on TechCrunch summed up Runway’s push as a bet that video intelligence can become a broader AI platform, not just a tool for directors, marketers, and visual effects teams. The company was founded in 2018 by Anastasis Germanidis, Cristóbal Valenzuela, and Alejandro Matamara-Ortiz, who met at New York University’s ITP program. It has since raised about $860 million, including a $315 million Series E in February led by General Atlantic with backing from Nvidia, AMD Ventures, Adobe Ventures, Fidelity, AllianceBernstein, Felicis and others. This round valued Runway at $5.3 billion.
These numbers are big for a creative software company. If the other party is Google, it will be smaller. But Runway also has an important point in a market where many AI labs are still burning through cash before proving demand. One founder told TechCrunch that the company increased annual recurring revenue by $40 million in the second quarter of 2026.
This revenue gives Runway a substantial advantage that many research institutions do not have. Its tools are used across film, advertising, marketing and digital content. Lionsgate announced a partnership with Runway in 2024 to build custom AI models around the studio’s film and TV library, but AMC Networks has since moved on to incorporate Runway’s tools into its marketing and TV development work.
This is not a minor detail. Paid users publicly provide Runway with feedback, distribution, and reasons to continue improving the product. The filmmaker doesn’t care if the model has an elegant internal expression. They care about whether the characters are consistent, whether the camera movements work, whether the physics are believable, and whether the clips can be used without wasting their day.
That pressure helps. Runway’s Gen-4.5 video model was released in December and has since been updated with native audio, long multi-shot generation, character consistency, and more advanced editing controls. In independent video benchmarks, it ranks higher than Google and OpenAI’s models in some rankings, helping to explain why investors are more willing to fund startups that take on larger research races.
But benchmarks are only part of the story. A video model that can create a convincing shot doesn’t automatically become a model that understands the world. They may have learned enough structure to imitate movement, lighting, and cause and effect. Or maybe it’s learning some sophisticated shortcuts. This distinction is important as Runway wants to move from production workflows to robotics, drug discovery, climate modeling, and gaming.
computing is a difficult question
Runway’s argument is that video is a natural path to intelligence because the world reaches us as a stream of visual, physical, and temporal signals. If the model can learn from it, it can potentially simulate better than systems trained primarily on text. This is the logic behind GWM-1, the first general world model with applications for interactive worlds, robotics, and avatars.
Google is taking a similar question from the opposite direction. DeepMind has spent years developing agents, games, robotics, and simulated environments. While Genie 3 works and Project Genie prototypes aim toward an interactive world, Veo 3.1 continues to push Google’s video stack into authoring tools like Flow, Vertex AI, and the Gemini API. Google also gives AI startups what they want most: direct access to huge infrastructure.
Runway has sought to close that gap through partnerships and capital. TechCrunch previously reported that the company had signed a deal with CoreWeave to expand its computing capacity, and that Nvidia and AMD’s participation in the funding round was strategically important. Still, buying or renting enough computing to train cutting-edge video and simulation models is a different burden for startups than it is for Alphabet.
That’s why Runway’s move beyond Hollywood is both ambitious and necessary. The film production market can support serious companies, but may not justify the cost of training the next generation of world models on their own. Robotic companies need a synthetic environment. Drug discovery teams need better simulation. Climate researchers need models that can reason about entire messy physical systems. Game studios want a world that responds in real time. If Runway can serve even a few of these markets, its valuation starts to look more like a platform bet than a creative AI premium.
The risk is that video intelligence does not generalize cleanly. Models can generate beautiful storms without understanding the climate. Even if it’s not reliable enough to train a real machine, you can still show a robotic arm lifting a cup. It can render molecules and laboratory scenes without being useful for drug discovery. The whole challenge is to make the leap from visual plausibility to useful predictions.
Runway doesn’t need to beat Google everywhere. Before the largest research institutions turn this category into another infrastructure contest, focused startups rooted in real-world creative workflows and driven by paying users need to prove they can build commercially useful world models. The next thing to watch is not just whether Runway’s videos look better, but whether its simulations start helping companies make decisions outside of the editing room.
Also read: Samsung’s AI chip boom turns labor into a supply risk • Bill Ackman bets Microsoft can overcome AI spending fears • Lake Tahoe power crunch shows AI’s hidden infrastructure costs
