Owning the Video AI Stack
Cristóbal Valenzuela, CEO of Runway, on the state of generative AI in video
Owning the stack lets Runway turn product usage into a research advantage, not just a software feature. Because it controls the model, the video pipeline, and the app where people actually edit, it can see exactly where creators get stuck, which outputs feel too slow or too brittle, and which workflows deserve model work versus simple UX fixes. That is especially valuable in video, where latency, rendering, and frame consistency are product problems as much as model problems.
-
Runway has long treated the rendering backend as mission critical and built it in house, while buying generic infrastructure around it. In practice, that means keeping control of the pixel level editing system that powers tools like rotoscoping and inpainting, because that is where product differentiation shows up for users.
-
This is the opposite of companies that add a thin product layer on top of third party models. Runway pairs proprietary video models with a web editor and filmmaker workflows, while horizontal labs like OpenAI bundle video inside broader consumer AI products, and wrappers like Pika or OpusClip focus on narrower jobs.
-
The payoff shows up in speed to production and monetization. Runway has emphasized moving models from research into usable tools quickly, and that full stack workflow helped it expand from editing automations into Gen-3 video generation, driving estimated ARR from $25M in 2023 to $84M in 2024.
As video AI gets cheaper and more capable, the winners are likely to look less like model vendors and more like full workflow owners. Runway is building toward a position where research, deployment, collaboration, and distribution sit in one loop, which should make its products improve faster as more professional video work moves into AI native tools.