YouTube Nears 30B Videos, But 99% Are Just AI Training Data

YouTube is about to hit 30 billion videos, a milestone driven almost entirely by Shorts, but a new report from Omdia reveals that 99% of its content gets almost no views, serving instead as a massive training ground for Google's AI.

• The 99% problem: The platform's viewership is incredibly concentrated, with just 1% of its videos capturing 91% of all watch time. Despite hosting the equivalent of 280,000 years of content, the vast majority of uploads are rarely, if ever, watched by human eyes.

• Data for the machine: That massive library of unwatched clips—the "long tail"—forms "the backbone of Google’s video training data for Gemini," according to Omdia Senior Analyst Daoud Jackson. While viral hits drive perception, Jackson notes "the reality is more complex," with the unwatched repository playing a key role in Google's AI strategy.

• What we're watching: When humans do tune in, they stick to the familiar. The vast majority of watch time (nearly 80%!) is spent on professionally produced content and music. News accounts for another 10%, while fast-growing video podcasts now capture 5%.

YouTube has effectively become two platforms in one: a highly concentrated entertainment service for its human audience, and a vast, largely unseen data farm to power Google's AI ambitions.