If you’re taking a similar route to YouTube, you also need a ton of CPU/GPU power and/or specialized hardware. YouTube transcodes every video into 2 (3 for videos with >~2M views) different formats in 5 different resolutions. A community-run service could skip on some of that, but it’d come at the cost of lower quality, less support for older devices, or higher bandwidth usage.
Out of curiosity, what does CPU/IO usage look like for SDF Chatter servers at the moment?