I implemented an encoding pipeline for AV1 for vids uploaded to my social news site (think reddit competitor except I'm extremely small fry). I eventually removed the code for it.

While the space savings and quality improvements are good, the encoding speed is an order of magnitude slower than using h264/vp9. In the end the user experience of causing people to wait significantly longer for an AV1 encode wasn't worth the tradeoff. To fix the user experience problem, I still had to encode a h264 version anyway, which kinda defeats the point when it comes to space savings. You still get data transfer improvements, but the break even point for when the encoding costs offset the data transfer costs were around 1000 views per min of video encoded, and as an average I'm far below that.

IMO there's a reason why YouTube only encodes AV1 for certain videos - I suspect it's based off of a view count. Past that point they trigger a AV1 encode, but it isn't worth it to do all videos, at least right now.

Worth keeping in mind I was looking at this ~2 years ago, so things may have evolved since then.

>IMO there's a reason why YouTube only encodes AV1 for certain videos - I suspect it's based off of a view count. Past that point they trigger a AV1 encode, but it isn't worth it to do all videos, at least right now.

But how can they do that without storing the original uploaded video until it hits that view count?

Do they actually store the original uploaded video somewhere, but reencode for the edge servers to save data/storage?

Things have gotten a lot better. You were probably using the reference encoder, but there's a newer, much faster encoder: svt-av1.