Yeah, mainframes are all about cases where the uptime is critical. Most modern systems are good with offering 99.9% or 99.99% reliability, with the understanding that trying to offer more than that just gets more and more expensive. Well... spending huge amounts of money on mainframes is one of the strategies to get to reliability numbers like 99.9999999%.

Source: https://itic-corp.com/itic-2023-reliability-survey-ibm-z-res...

The legitimate usage is typically for workloads like relational databases (which are anyway single-machine architectures) that experience heavy load 24/7, part of fragile architectures that cannot tolerate downtime (remaining fragile as such because the original source code has been lost etc.), where "the system is down" causes tens of thousands if not hundreds of thousands of people to stop work.