Netflix serves over 40,000 hours of video in more than 100 encodings. Source files are high quality and can reach terabytes per hour of bitrate. Source data must be retained for re-encoding. Encoding is adaptive to scene. The ETL can generate a quarter million tasks to process one source file.

I’m not saying YouTube’s problem isn’t much larger, but your assumptions are off by orders of magnitude. If you are curious about the workload there are many blog articles and stories here.