Not only has this idea been around for a while as he says at the beginning, it has been implemented a couple times over already. And not just the theoretical version he did that doesn't work because of compression, but variants that try to squeeze out the maximum, add error correction, use audio too etc.

a few ones pop right up when you search things like "using YouTube as storage" but thanks to YouTube search being useless garbage nowadays I can't find the particular one I'm thinking of. It was rather long but really went into the technical details and challenges.