> The largest cost by far would be bandwidth, storage space is insignificant, because the likelyhood that any relay would have to store everything is essentially zero (and that's a good thing).

The largest cost by far would not be bandwidth or storage but rather the cost to index these relays. By several orders of magnitude.

Nostr as it stands cannot scale to millions of users without indexing—or put another way as it scales power and profit will naturally accrue to indexing parties to the degree that Nostr will lose itself. And since indexing (naturally) isn't part of the protocol, it will be done in a computationally inefficient manner, with multiple competing parties performing the same tasks, and then falling off one by one due to the immense costs each party must bear alone due to the lack of cost-sharing.

This will become much more obvious after around 500k daily active users, should Nostr one day hit that mark. It's not an issue now, but the initial hints of it are here.

There are only two ways to avoid the more-scale-more-indexing trap. The first is to stick to use cases that make indexing irrelevant (it doesn't help with anything). The second is to accept that core use cases will be those for which indexing does bring about benefit at scale but bake indexing into the protocol at the most fundamental level. Both are still roads Nostr could take.

And there is a deus ex machina that some kind of decentralised indexing technology comes along and is simple and just works.