Nvidia caught scraping YouTube and Netflix to train AI

Nvidia has been scraping massive amounts of video data from YouTube, Netflix, and other platforms to train foundation AI models, according to leaked internal Slack chats, emails, and documents obtained by 404 Media.

Internal documentation described the aggressive collection effort as scraping roughly "a human lifetime" of video every single day.

A former employee who spoke anonymously to 404 Media revealed that the scraped footage fed an unreleased project internally dubbed Cosmos. Distinct from Nvidia's existing deep learning software of the same name, this initiative is building models for the Omniverse 3D world generator, autonomous vehicle systems, and digital human avatars.

When employees raised red flags about the legal risks of tapping YouTube clips and datasets compiled by academics for research, managers claimed the practice had clearance from the highest levels of the company. Nvidia defended itself by stating its actions are "in full compliance with the letter and the spirit of copyright law."

Why it matters

Even the world's most valuable chipmaker cannot escape the messy realities of generative AI training: frontier multimodal models are starving for video data, and public platforms remain the easiest buffet.

Seeing Nvidia target YouTube and Netflix mirrors the controversies previously surrounding OpenAI, showing that hardware vendors are just as aggressive about hoarding media as software labs. Telling staff they have internal executive clearance does not resolve external copyright infringement questions; it simply shows how willing tech giants are to ask for forgiveness later.

For creators and streaming services, this proves once again that terms of service and paywalls do very little to prevent proprietary creative work from becoming free raw material for enterprise AI.