Ok, since you've tagged Mike. I think we need to change into English discussion. 😅

Mike has already give us a bit answer below.

I just want to add a bit,

> How we can scale on later? Let's say there are 100 million users using Nostr. How much data storage later that we need?

If we implement NIP-65 in many clients we can reduce the burden of relays. Since not all relays have to store all our nostr events. Many relays just need to store kind 0, and 10002 for discovery. Those data are very small between 0.5-2 KB. Assume 1 KB of metdata event, we just need to store around 100 GB to store the latest event for kind 10002 for 100 million people as the example. It is quite easy to store that in the server.

Since users can use different relays, the data (notes, likes, zaps) doesn't have to be stored in all relays thus reducing the burden of relays.

In clients, we just need to cache profile data (kind 0) and relay list (kind 10002) of our follows (small maybe we follows 1000 people). Smart client will fetch data only when we need it saving the bandwidth.