Earlier quoted context omitted.
you really should scale vertically until you can't anymore. a server with two 64 core epyc CPUs and 3TB of RAM and 24 NVMe flash SSDs is going to be able to handle 99% of companies SQL DB needs.
I mostly agree with you, but a single server won’t solve redundancy and disaster recovery. That doesn’t mean you need to adopt a fully distributed system – a read replica or even periodic backup should be sufficient – but it’s not as simple as “just use a single server.” Stackoverflow is famously powered by a cluster of three vertically scaled database servers.
Like you say, deploy read replicas or automate regular backups (with regularly tested restore automation), whatever you need to meet your RTO and RPOs.
You need to be able to ignore any ignorant sales/marketing/growth management that try to tell you "we can't afford any downtime at all!!!" and have enough internal political power on the tech side. Tech leadership should be able to document the costs of each extra nine of uptime to senior management and have them acknowledge that occasional second or perhaps even minutes of downtime during failover is acceptable.
By the time you start approaching the limits of your server with two 64 core epyc CPUs and 3TB of RAM and 24 NVMe flash SSDs - hopefully you no longer need to take advice from randoms on HN because you should by then have 100+ database and network engineers working follow the sun shifts with some _really_ smart and deeply experienced dedicated database leadership managing them.
Or, you need to get someone on board to teach all your junior devs or "vibe coders" about database indexes and how to construct queries that don't do multiple fulltable scans just to render some user profile widget on every page load... "But it wrks fine on my laptop!!!" (with a whole 7 user accounts and 47 rows in the user_activity table...)