Because nowadays you can get servers with about 192GB of RAM and 32 CPU cores.
Clustering seems like one of those "what if"[0] scenarios where maybe if you were operating at roflcopter scale you might need it but for 99.9999999999999% of cases, a single master DB is more than enough -- at least with a well designed SQL db like Postgres.
[0]: https://nickjanetakis.com/blog/optimize-your-programming-dec...
With a load balancer, you can easily patch your web servers, but very often, DB servers go unpatched and un-upgraded for years since they aren't in a self-healing cluster.
To add some detail, I worked on an e-commerce system that was like this. We would've lost something like $60k an hour if we rebooted the database server (yes, one server, nobody expected this sort of growth). So we didn't touch it for 3 years until it was about to run out of disk space. Then we had no choice, but then we set it up as a fully replicated system with a ton of disk space, making future patching easier.
A 30-second outage to switch to a replica is perfectly fine for production.
You can get servers with 2T of memory and 96 cores off the shelf for a few years now. Real world workloads that you can't run on these boxes are few and far between. But everyone worries about "scalability" like they'll ever run into those limits.
I can’t imagine any company crazy enough to run one server like you’re describing. Definitely wouldn’t pass any enterprise service continuity testing/review that’s for sure.