The volume and storage numbers you quoted call for breaking up a monolithic system. We have a similar overall volume of transactions and storage needs to yours but we reside in all US regions, tens of AZs. If one AZ went down, a small fraction of customers would be affected. True, we have increased our exposure to an outage happening somewhere but we improved overall business continuity. Simply put, dealing with 5% of pissed customers is easier than 50% or all of them calling you at once.