My assumption was based on the idea to have a proper YouTube replacement. Not some run down video storage for a hand full of large content creators that can afford it.

• The scalability you buy via P2P also means an increased storage. So if you want to offer a similar platform that is used in a similar way then you probably would need a multiple of the current storage capacity that YouTube offers. Likely close to an exabyte of storage (assuming that YouTube has just about 300 petabytes. Which likely is a lower number by now.)
• Especially for the amount of users consuming the content you would need a good distribution factor. Popular content would need to be distribution over thousands of peers for it to kinda work out. So a lot of people could share the necessary video data, making the storage a problem.
• Big servers in a datacenter will always be more efficient because they are designed to be compared to consumer hardware. It's like replacing a central power plant with a small power plant per home. It won't deliver the same efficiency and is a waste of resources. Ecologically speaking.

creators already store their content locally

A lot of creators delete at least the raw footage because they don't have enough space and it would be too expensive. One creator hosting their own content wouldn't even begin to scale in such a scenario. They would need powerful hardware and serious network connectivity. Something the large creators probably could afford, but most couldn't.

peertube can run on rather old tech so I’d say it’s more efficient.

Especially old tech is less efficient than current generations.

tl;dr: I think you were talking about a small solution for large content creators where as I took it as a literal replacement for YouTube.

The Internet is not a "series of tubes" ... It's a packet-switched messaging network. The fact that billions of computers are "connected" to a single address doesn't really mean much other than they've exchanged some messages within the last several minutes (or some other arbitrary amount of time).

You're not wrong: any sizeable web service must distribute to several servers and data centers for performance (e.g. response times and data throughput), and for resiliency (e.g. if a server fails then another one can take over). But the difference is these data centers have a financial incentive to maximize efficiency in both hardware costs and electricity usage (which includes cooling, etc.). Folks self-hosting Lemmy/Mastodon/etc. servers in their basement have much less incentive, and so less effort is put into eeking out every ounce of capability per dollar. Even hosting on AWS/Google/Azure/etc is never going to beat a bespoke data center dedicated to one particular application.

Although they don't necessarily publish this information, at least a data center can accurately measure its energy usage (which tends to dwarf hardware costs...). Also newer hardware will always outperform old hardware per energy usage. For either aspect I can't say the same for the server in my basement ... It's 10 year-old hardware running on the same circuit as the beer fridge next to it. I have no idea how much electricity it uses to handle like 2 users. It's a glorified space heater.

It's all about trade-offs. Fediverse applications value open standardization, availability, and long-term resiliency over efficiency, performance, and short-term profits.

The Fediverse is great, but in the short/mid-term, efficiency and ecological impact aren't things i would expect it to excel at.