[bluGill]

Buffering and waiting for TCP slow-start to get up to speed. While the amount of buffer time needed can be debated, given that in a long stream it is likely that a packet will be lost and resent via normal TCP (or just sent via a different route and so arrive late), you should have a few seconds buffer, and that means a few seconds at the start before the stream starts. It is a technical thing.

You don't have to implement things as above. For live video you probably should use UDP and design your protocol so that you can handle a few missed packets - that is have your video become fuzzy in those cases. This is a lot more complex to design though and so not would you should do first. Even if you have Google's engineers, the first solution is better for not-live video since a few more seconds of delay mean you can keep clear video.