Eight years after Cristóbal Valenzuela started Runway in a small New York studio, the company is now valued at $5.3 billion.
But Runway no longer wants to be called an "AI video company." In an interview last Friday, co-CEO Anastasis Germanidis made a striking statement:
"We're basically bound by our own understanding of reality. Language models are trained on the entire internet, but to get beyond that, we need to leverage less biased data."
In plain terms: large language models can only learn from what humans have written online, which has reached its limit. To truly understand the world, AI needs to watch video and learn on its own.
The company's numbers
| Metric | Data |
|---|---|
| Latest valuation | $5.3B |
| Total funding | $860M |
| Latest round | $315M (Feb 2026) |
| Q2 new ARR | $40M |
| Employees | 155 |
| Offices | NYC, London, SF, Seattle, Tel Aviv, Tokyo |
Adding $40 million in ARR in a single quarter is healthy growth by SaaS standards, but not aggressive among AI companies — Cursor went from $100 million to $2 billion in a year. Runway is clearly playing a longer game.
The world model path
Runway released its first world model last December and plans another this year.
What is a world model? Simply put: given a video, it predicts whether a cup on the table will fall, how a person will move, and how light will change in the next second. It learns physics, not just "looks like video."
The difference between this and video generation is like drawing an apple versus knowing an apple will break when it hits the ground.
Competitors on the same track
Runway is not alone:
- Google: Veo video model + Genie world model, both in development
- Luma AI: $900M total funding
- World Labs: Fei-Fei Li's company, raised $1.29B
- OpenAI: Still pursuing the Sora path
None lack funding, but the difference will come from who first makes a world model that works. Germanidis believes video is the cheapest and most direct training source. Rather than training on every math problem or philosophical debate, letting the model watch camera data gets closer to reality.
What Runway is betting on
If world models succeed, they could unlock several industries:
- Robotics: Let robotic arms anticipate what happens next
- Drug discovery: Simulate molecular movements
- Climate modeling: No need for humans to write formulas
Each is far bigger than "making a TikTok filter."
But there is a problem. Video datasets are more complex in copyright and scale than text — you can crawl the entire internet for text, but video? Copyrighted content on YouTube is already fueling lawsuits against Google and OpenAI. Runway's press release did not mention how it will source data.
A $5.3 billion valuation is not cheap, but compared to World Labs' $10 billion and Cursor's $50 billion, Runway is relatively modest in this race.
Who will emerge? It depends on which company turns a "world model" into a profitable product in the next two years.
Sources: CocoLoop, Runway started by helping filmmakers — now it wants to beat Google at AI (TechCrunch)