Most outages are not regional

A design that survives losing a region is prepared for one kind of failure, and it is not the most common kind. Google's SRE book attributes roughly 70% of outages to changes in a live system, and a change can reach every region at once, because every region runs the same code and reads the same configuration.

Cloudflare's July 2019 outage is the clearest case. A single firewall rule with a badly behaved regular expression went global in seconds and pushed the CPUs serving HTTP and HTTPS to nearly 100% across the network. Many locations did not help; all received the rule. At the Financial Times, a DNS fault took out every ft.com name, including the dashboards and runbooks for diagnosing it, a reminder of The dependency you did not list.

Regional outages still happen; AWS isolates its regions so that one can fail while the others keep running. The better question is which failures a design survives, and whether the effort went to the likely ones. Defending against global changes looks different: Staged rollouts contain a bad change, and copies that are not updated in lock-step (Replication copies your mistakes too).

It is worth asking of the original claim first.