Im very skeptic about the scalability of the current architecture.
Imagine 1 billion people using bitcoin 5 times/day = 5 billion transactions / day = 1825 billion transactions per year
Is the bitcoin client able to handle 58,000 transactions per second and a blockchain terrabytes or maybe even petabytes in size?
OH NOES Bitcoin might only someday handle 10% of all tx in the world.
Come on nobody now how large Bitcoin will be. Maybe Bitcoin never gets to PayPal stages so worrying about will it be able to handle being the single one currency in the world where every man woman and child processes every single tx is kinda silly. I mean Bitcoin isn't even the single cryptocurreny today and your are invisioning a system where every nation on earth gives up their sovereign currency (something that is almost inconceivable) and the entirety of humanity processes directly on the blockchain for everything from billion dollar (equivelent) intra bank swaps to a cup of coffee.
If Bitcoin can't do that but it can say become the world largest payment network that is open, transparent, and decentralized in a decade or two would you consider it a failure?
Still if you absolutely can't sleep without knowing can Bitcoin scale to "the one coin to rule them all":
1) Will all tx be on the blockchain? It is highly unlikely given that even today with essentially free txs and a tiny tiny ecosystem all tx aren't on the blockchain.
2) When will this end game occur? Tomorrow, a decade, fifty years?
Today handling that type of volume would require very large and expensive nodes. It probably could be done but we are talking some serious hardware. The largest bottleneck is bandwidth and most residential connections don't come close so all full nodes would be in a datacenter. To avoid near continual swapping one would want a significant fraction of the UXTO in memory and at 50,000 tps we are talking "a lot" (64GB? 128GB?). Disk is less of an issue but you are talking about ~1 TB per day. It would require some pretty massive RAIDs. Possible but hardly hobbyist. CPU power is almost a non-issue today (quad core can perform roughly 50,000 tps) and will become less of an issue in the future.
The good news is Bitcoin isn't going from <1 tps to 50,000 tps in a year, or a decade. It may never scale that large for issues beyond tech but even if it did we are talking on a lifetime scale. Today we have TB scale drives, when I went to college in 1995 I bought one of the first economical 1 GB HDD. I have do doubt in another 2 decades we will have PB scale storage for $200 or so. The same will apply to bandwidth and memory.
It probably makes more sense in looking at "if in the next 5-10 years" Bitcoin reached PayPal scale (tens of million users, 50 tps) what kind of resources are we talking about. What types of optimizations would pay the largest dividends.