Ok. Here goes. The Steem blockchain has a three-second block processing time target. What this means is that when/if its user base grows many orders of magnitude, the blocks will have to also grow. Some people complain that it will force witnesses to upgrade their hardware to such an extent that running a witness node let alone an RPC node will be an industrial operation. My comment to that is that so be it. Steem is built to have only 20 witnesses who participate in block production anyway. So I don't get what the problem is. Besides, the price of memory, bandwidth and storage are dropping exponentially anyway. Since Steem does not store binary files, only text, I don't see scalability as an insurmountable obstacle.
My topic suggestion is scalability. Steemit Inc is working on Multi Indexing RocksDB Adapter. The idea is to move away from storing the blockchain in RAM and to store it on disk (RocksDB). It would be interesting to hear estimates on the costs of running a witness node or an RPC node after the transition to RocksDB is complete. How much will it help?