I'm starting to understand the architecture more (I think), and am feeling more confident about the long-term future than I was, but here are some of my thoughts about what you could do:

1. Add some differentiation to the RPC server:
   a) Allow it to be run to provide only the last n days worth of vote, comment and transfer data. This is all I would need for running my services on it, and I expect many others could also manage with less historic data. A smaller index surely reduces the RAM requirements significantly.
   b) Allow it to run in a 'ledger only' mode, where the index could be smaller (as it would exclude all other operations). This is almost all the exchanges would need I think.
2. Communicate better about how the infrastructure works and the hardware requirements for running each part. I'd love to see some diagrams of the systems architecture.

I realise this will need some work on steemd, but over time, these increasingly demanding running requirements are centralising the network unnecessarily, making it more vulnerable, stifling community innovation and leading to doubts about scalability.

Thanks for your time.