Unit Economics in AI Video
Cristóbal Valenzuela, CEO of Runway, on the state of generative AI in video
The hard part in AI video is not just making a good model, it is making every generated second cheap enough to sell at consumer software prices. Text models return tokens, but video systems must create and move many frames, then handle playback, streaming, and editing responsiveness. That is why Runway built its own full stack, because margin, latency, and creative control are all tied together in video.
-
Runway prices like SaaS, not like raw compute. The product has been sold from about $12 to $76 per month, while management has emphasized charging for creative value rather than billing users for every GPU cycle. That only works if infrastructure is heavily optimized behind the scenes.
-
Video jobs are more varied than text prompts. A tool might need to rotoscope a subject, replace a background, expand a frame, or generate a new shot with consistent motion. Each task has different compute and latency needs, so one general cost structure is harder to achieve than in text.
-
The payoff is large when the economics work. Runway has been used to cut tedious VFX work from hours to minutes and reduce cost per shot from about $350 to about $10. That makes small teams viable in workflows that used to require large specialist crews.
As model efficiency improves, the winners in AI video will be the companies that combine better generation with cheaper delivery and faster editing loops. That shifts the market toward vertically integrated platforms like Runway, and away from thin wrappers that can demo video generation but struggle to support sustained, high frequency creative work at scale.