Relays are going to have to start calculating a 'value score' for events and pubkeys. 

Only the events and pubkeys with value will survive being dropped - unless perhaps you pay a relay for persistence. What is value? How best to calculate it? Access frequency/recency? Keeping which events give relay users the most value? How could the value score be gamed/abused? We don't know exactly yet - and every relay using the same 'value' scoring just means relays all store the same higher 'value' data... aka. convergance + centralisation.

Why? Well, managing databases in the 100s GB is not just start and forget. Cheaper cloud servers often have a base disk of around 100GB. Relays need to avoid being abused as long term storage of misc data. Within six months we could start to hit some of these growth challenges for relays.

This will especially be important for the smaller relay ecosystem, and to enable smaller relays to continue to exist and be easy enough to operate. Even with a pay-to-relay gate, a malicious user could covertly take up GBs of DB storage with custom event data. Maybe relay paid users pick a storage plan with max GB? Maybe only accept certain event kinds? Again, how do we decide what to keep and what to drop..?