That only solves maybe one of the listen problems. Whatever instance you have, you still have to get and serve media to other viewers and instances. The only problem that this solves is potentially CSAM spam/moderation.

Let's say it was a cell phone, it could handle maybe 2 concurrent transcoding streams before stalling out and people running into buffer times (which makes them leave).

If every person had their own tiny, low powered servers, then you could have max like 5 concurrent transcodes on any instance in all of peertube for old laptop or desktop computers. Assuming an average of people have a 100/30Mbps connection (which is true in much of the world outside of major cities, or even lower), then that would be absolutely maxing out at 10 concurrent viewers if everyone is running AV1 compatible clients (which is not the case) and more like 6 concurrent viewers per video at h.264. Those estimates are at low bitrates also, so low quality, absolutely no slowdown from your ISP, and absolutely no other general home or work-from-home use. In reality it would be closer to 3-6 concurrent viewers per instance (not even per video)

Still not even counting storage which is massive for anyone that creates more than a couple videos per year.

My point is just that it is an extremely difficult and costly problem that is not as simple as "more federation" like in text and image-based social media because of the nature of video, the internet, and viral video culture. Remember, federation replicates all viewed and subscribed content on the instance (so the home instance has to serve the data and both instances have to store it)