The Engineering discipline relies on empirical studies. With constantly full blocks, you have no way to measure transaction demand.
Fees are only a weak indicator of transaction demand because of substitute goods.
Flowee the Hub (Bitcoin Cash full node implementation) can sync the equivalent of 4GB blocks using just a cheap quad-core VPS.
But being able to barely keep up means that if you fall behind for even a moment (say, if your connection drops... or if miners get lucky and mine a bunch of extra blocks) then you will _never_ catch up. Operating anywhere near the limit of your processing rate is non-viable for that reason.
You need to be able to process many times the network's capacity so that you can catch up in a reasonable amount of time.
To get back to reality, I don't think any sane person has advocated for anything close to a 2GB blocksize cap at any point in the near future. Bitcoin Core uses a maximum 4MB block weight and Bitcoin Cash uses a 32MB cap currently. This cheap VPS could sync the entire Bitcoin (Core) blockchain from 2009 to present in less than 12 hours. The biggest block size cap (not full blocks, the cap) being tested seriously is 1GB. I'm sure you know all this, though.