First, a global system with no outages (say the gold standard of 99.999% availability) is a promise which is basically impossible to keep. Second, a global sys

First, a global system with no outages (say the gold standard of 99.999% availability) is a promise which is basically impossible to keep. Second, a global system being always available definitely doesn't mean it is always available everywhere. A single datacenter or even a larger region will experience both outages and network splits. It means that whatever you design on top of the super-available global system will have to deal with the global system being unavailable anyway. TLDR is that the clients will have to tolerate outages (or at least frequent cut offs from the "global" state") anyway so it's better not to give them false promises.