Reading time 6 minutes
Tags Runway AI Video Generation World Model Generation AI Robotics Drug Discovery Computing AI Strategy
Runway, an AI video generation startup, doesn’t have a typical Silicon Valley pedigree. It doesn’t have a Stanford University founder. It doesn’t have a former Google founder. It doesn’t have a nine-figure seed round that bought time by ignoring revenue. Of the three founders, two are from Chile and one is from Greece. They met at New York University’s Tisch School of the Arts and founded the company in New York.
Runway may also depend on who you ask who is one of today’s most important AI companies It’s not about what they’ve built, it’s about what they’re going to build next
core bet
For the past few years, the AI industry has largely operated on the premise that intelligence resides in language. Large-scale language models like OpenAI’s ChatGPT and Anthropic’s Claude reflect that bet.
Along with other competitors, Runway is building something different.The founders believe that the next form of AI intelligence will be built not from text, but from videos and world models that learn not just how humans describe the world, but also how the world works.That distinction sounds academic, but its implications are not.
Anastasis Germanidis, co-founder and co-CEO of Runway, said training models directly on the world’s observed data is the next frontier in AI. He argues that the first companies to get there won’t be the ones that perfect the language.
We are fundamentally bound by our own understanding of reality. Germanidis spoke to TechCrunch from Runway’s homey, sun-drenched headquarters near Union Square.
Language models are trained by extracting existing human knowledge from across the internet, on message boards and social media, and in textbooks, Germanidis continued, but to go beyond that, they need to leverage less biased data.
From video tools to world models
Founded in 2018, Runway has built a reputation for video generation models such as its latest Gen 45 and AI tools that can transform text prompts into editable, cinematic content.
Today, Runway’s technology powers the production workflows of filmmakers and advertising agencies, and the company has signed deals with major media players like Lionsgate and AMC Networks. The tool has also been used in films such as Everything Everywhere All At Once.
Runway is now valued at $53 billion, and one of its founders says it will increase annual recurring revenue by $40 million in the second quarter of 2026.
If Runway bets that video generation is the path to the modeling world, the results will be felt from Hollywood to drug discovery, but otherwise Runway risks being overtaken by competitors with far deeper pockets, including the head of Google.
Within the past six months, the startup has put its plans into action and expanded beyond video generation. The company announced its first global model in December and plans to launch yet another global model this year. A world model is an AI system that simulates the environment enough to predict how it will behave.
Runway is not alone in its pursuit of turning physics-aware video models into world models, with short-term use cases in interactive entertainment games and robotics training. Startups Luma and World Labs are on a similar trajectory, and Google is pointing its Genie world model in the same direction.
Everyone wants some version of the same AI that solves humanity’s toughest problems. It’s a far cry from the original product on the runway, but it’s the result of both the technology’s new capabilities and its founders’ inclination to follow where it leads.
Why is the world model important?
Germanidis sees the world model as a scientific infrastructure: The more sensory data and observations that train a single model, the closer we get to a working digital twin of the universe, and the faster we can run experiments than any other laboratory. Much of the scientific process, he points out, is just waiting for results. If we can compress waiting, we can compress progress itself.
If we can train scientists to be better than human scientists, we can accelerate progress in how we understand the universe and solve problems, Germanidis said.
The kinds of problems that have puzzled researchers for decades, such as robotic drug discovery and climate modeling, are the kinds of problems Runway launched last year in its robotics division, which Germanidis says is already leading to real-world testing and deployment.
Gellman, like others, believes the field is moving toward training a single model with different modalities, text, video, audio, and other sensors, and believes compounding effects are important.
Given enough time and resources, his own runway technology moonshot goals are biological world models and anti-aging research.
resources and competition
Whether Runway can bring its video advantage to WorldModel is far from over, and the competition isn’t waiting Runway is one of the first companies to develop AI video generation, but WorldModel is a different breed with well-funded and well-regarded competitors Yann LeCun, former meta chief scientist at Google Fei Li, godmother of AI, and the growing startup field are all chasing the same goal
Kian Catanforouche, CEO of AI skills benchmarking company Workera and a lecturer at Stanford University, noted that while no one has yet proven the jump between video intelligence and generalized inference via world models, it’s not impossible, and said Runway needs to continue to marshal the resources of computing chiefs among themselves if they want to make their world model bets a reality.
Runway has contracts with CoreWeave and Nvidia, but has not confirmed whether they have dedicated cluster access that would guarantee the large-scale compute needed to train Frontier models.
How do you plan to build a basic model without using clusters, asked Catanforouch? I don’t think anyone can do that
Runway has raised $860 million to date, including a $315 million round in February from strategic partners including AMD Ventures and Nvidia. That’s about the same as its closest competitors, Luma AI and World Labs, which raised $900 million and $129 billion, respectively, according to PitchBook.
But Runway is also up against incumbents such as OpenAI, which raised about $175 billion per CEO Sam Altman, and tech giant Google, whose parent company Alphabet is worth $486 trillion.Google is Runway’s biggest threat.The company’s Veo model competes directly with Runway’s video generation business, while its Genie World model targets the same long-term space Runway is aiming for.
Catanforouche gave a nod to the fact that OpenAI shut down its video platform Sora in March after consuming about 1 million in computing costs per day and by some estimates reaching just 21 million in revenue. He noted that resources alone do not guarantee survival, and neither does Runway.
culture and execution
Catanforouche isn’t writing ‘Runway’ He argued that ElevenLabs, an AI audio startup that outperformed OpenAI and Google in its own benchmarks, could follow a similar strategy, despite lacking the resources and pedigree on either runway.
Valenzuela, Runway’s founder, says the startup’s lack of standardization in the Bay Area gives it an advantage. Not only do they have a diverse range of ideas, as he argues, but without their Silicon Valley connections, without the war chest that many of their peers have access to, Valenzuela says they would have been forced to be more scrappy and removed from the need to generate revenue early.
And Michelle Kwon, Runway’s chief operating officer, said the company is in no hurry to raise more funding, even as computing demands increase with scale.
Their backgrounds have allowed them to build a culture that often makes the right decisions early and moves incredibly quickly, Compound’s early investor Michael Dempsey, managing partner, told TechCrunch.
For Valenzuela, that culture begins with how we see the world in the first place. He is a co-CEO and spends his limited free time as a new father reading books such as Chilean poet Nicanor Parra, who he describes as the antithesis of Pablo Neruda. Less formal and academic, he takes the view that poetry belongs to the people, not the rules.
Rules are just rules they invented, Valenzuela said. That’s what drives us to do things on the runway. They say Silicon Valley is here and that’s where startups are. Why are they just rules that were made? Erase them all and start again.
Do you think video-trained world models will surpass language models as the path to general intelligence, or do you think text will still remain the primary foundation? Share your opinion in the comments
