The part that doesn't scale is not delivery (which can be parallelized independently among nodes), it is the custodial directory function -the tracking, location, identification and availability of high quality seeds and hosts. This is the function that will start out optimistically decentralized by hobbyists and enthusiasts and eventually succumb to opportunistic centralization as anything close to serious scale begins to kick in.
YouTube as a service has very different performance expectations than something like bittorrent. for one thing the directory function is built in and not externalized to another set of directory services, which are notoriously unreliable in the case of BitTorrent anyway.
And unlike BitTorrent or Mastodon I don't only want to know where I can find a given file, I want to know where I can find it delivered immediately, consistently, smoothly, without hesitation, in a variety of cross encoded formats I'm almost certain to have supported on my device, and with some sort of consensus mechanism to establish a better than chance confidence that the file I start streaming is the file I expected to stream and not something else with the same filename.
YouTube as a service has very different performance expectations than something like bittorrent. for one thing the directory function is built in and not externalized to another set of directory services, which are notoriously unreliable in the case of BitTorrent anyway.
And unlike BitTorrent or Mastodon I don't only want to know where I can find a given file, I want to know where I can find it delivered immediately, consistently, smoothly, without hesitation, in a variety of cross encoded formats I'm almost certain to have supported on my device, and with some sort of consensus mechanism to establish a better than chance confidence that the file I start streaming is the file I expected to stream and not something else with the same filename.