OpenAI, Google and other AI giants have discovered a new source of training data — they are purchasing unpublished videos directly from content creators.
According to Bloomberg, companies are paying between $1 and $4 per minute, significantly more for high-quality 4K footage, drone videos, and 3D animations. Unused content from platforms like YouTube or TikTok typically costs $1–$2 per minute. Such deals often reach thousands of dollars, Bloomberg reports.
To manage licensing logistics, companies are collaborating with specialized firms like Troveo AI and Calliope Networks. Troveo’s CEO, Marty Pezis, claims that every business developing video models either already partners with them or is in negotiations — and they have paid creators over $5 million.
Dan Levitt from Wasserman describes the current landscape as an arms race in video content. He sees lucrative licensing opportunities arising in the coming years, but warns that they won’t remain available forever.
Contracts include protective measures that prevent AI companies from creating digital copies of the authors, recreating specific scenes from their channels, or using footage in a way that could harm the authors’ reputations.
Alongside purchasing footage directly, Google has recently introduced another option: they have added new features on YouTube that allow creators to control whether AI companies can use their public videos for training.
There is a list of potential licensees including over 17 companies, such as OpenAI, Meta*, and Microsoft. Since YouTube prohibits unauthorized content harvesting, these companies may serve as intermediaries and profit from future licensing agreements.
AI companies are eager to acquire all this video data to build their video generators, like Sora and Veo. However, some researchers have even grander ambitions: they believe that these massive video datasets could help create a sort of world simulator.
The theory, although scientifically controversial, suggests that watching a large amount of video can assist AI systems in better understanding physical reality and enhancing their generalization abilities.
This has become an unexpected windfall for content creators. Many filmmakers produce hundreds of hours of video footage each year that never make it online and previously had no value. Now they can sell this raw content for real money.
Source