I don't understand how ACID (atomicity, consistency, isolation, and durability) critical systems, like banks, can agree and decide what my account balance looks like at any given moment? Or is my bank account actually just a single SQL database row somewhere with no read/write replicas?
They do not. They have high availability systems and error correction/ledger balancing mechanisms. For e.g., a system partitioned by account numbers, generally works very well since the number of updates to an individual account is low and most of the updates are batched for end of day processing.