HW – if all you do is rack & stack hardware, you can manage thousands of boxes. You see this at Facebook/Google type scales.
OS – when you’re responsible for troubleshooting operating system problems (as opposed to simply re-imaging the box from scratch), you can manage hundreds of boxes or VMs.
SQL – if you have to install, configure, and troubleshoot standalone instances of SQL Server, you can manage 50-100 instances.
HA/DR – when you have to design, build, and troubleshoot high availability features like clustering, Availability Groups, replication, etc, then you can manage 10-50 SQL instances. (Here, it’s important to know that we’re talking about boxes critical enough to have a high availability solution.)
PERF – if you have to tune queries and indexes for performance, you can do a great job on 1-5 instances of SQL Server. Doing a good job of performance tuning means understanding the underlying data model of the databases you’re tuning, plus having a good relationship with the application developers.
These numbers aren’t absolute: tools like training and automation can raise them, and a lack of those tools can make matters worse.