(Edit: Though that may not actually work considering this is apparently fully server side. Gonna have to get clever…)

Next step is machine learning to recognize ads and cut them out automatically hah.

"Line go up" is the animating force of the age, the critical philosophical principal around which our entire society is arranged.

Gives me a fucking headache.

If you don’t pay for something, you are not a customer, you are the product. If you pay for Youtube, you don’t see the ads, but you are also still their product. Lose /Lose

Line go up 💹

This isn't how YouTube has streamed videos for many, many years.

Most video and live streams work by serving a sequence of small self-contained video files (often in the 1-5s range). Sometimes audio is also separate files (avoids duplication as you often use the same audio for all video qualities as well as enables audio-only streaming). This is done for a few reasons but primarily to allow quite seamless switching between quality levels on-the-fly.

Inserting ads in a stream like this is trivial. You just add a few ad chunks between the regular video chunks. The only real complication is that the ad needs to start at a chunk boundary. (And if you want it to be hard to detect you probably want the length of the ad to be a multiple of the regular chunk size). There is no re-encoding or other processing required at all. Just update the "playlist" (the list of chunks in the video) and the player will play the ad without knowing that it is "different" from the rest of the chunks.

It wouldn't cost any CPU with custom software that Google can afford to write. The video is streamed by delivering blocks of data from drives where the data isn't contiguous. It's split across multiple drives on multiple servers. Video files are made of key frames and P frames and B in between the key frames. Splicing at key frames need no processing. The video server when sending the next block only needs a change to send blocks based on key frames. It can then inject ads without any CPU overhead.

I mean placement within the video timeline. E.g. do all users see the ad at 0:00 or 2:00 or does it jump around for everyone to prevent it from being tagged.

This breaks the current SB implementation, but if the ad duration is known and consistent across the userbase then it will fix itself as users tag videos with the "new" timestamps.

Affected*

Storage more likely. Google owns fiber backbones and peers against the tier 1 providers directly. The over all point of 'no, it's still prohibitively expensive' stands unless you've got 20B of dark fiber in your pocket.