Your quantified cost comparison is the right way to decide. The VPS cost rarely includes the ops burden.
> Biggest win? Access controls.
Be careful scaling those tag-based ACLs. They're great for a few groups, but the JSON policy becomes a management headache past a certain complexity. Start documenting your access matrix in a flat table now, before you need it.
For subnet routers, keep it to one or two. They're a bridge, not a design goal. If you need more, reconsider why that infrastructure isn't in a cloud VPC with a proper gateway.
Data over opinions
The locked-down corporate device scenario is a great example. We've seen that exact TUN/TAP block cause the 45-minute session to balloon, because you're now troubleshooting group policy inheritance instead of just VPN settings. That's a silent cost multiplier most ROI models miss.
The spiderweb of subnet routers is a real architecture smell. We set a hard limit of three. If a fourth is requested, it triggers a mandatory review to migrate the underlying service or build a proper ingress, because you're just recreating the DMZ you were trying to avoid.
On the privacy point, opt-in is key. We also had to add a clear visual indicator in the client UI when the exit node is active, because some team members didn't realize their traffic was being routed.
Your points on setup and access controls are valid, but I need to push back on treating tags as a long-term solution for a 30-person company. They work until your first compliance audit.
That "easy" segmentation can backfire. Without logging who *approved* the tag assignment, you can't prove who had access to what and when. Tailscale's admin audit log is limited compared to a syslog server you control. For SOC 2, that's a finding.
If you're documenting the access matrix in a spreadsheet, add an "approved by" and "date" column now. You'll thank me later.
Where is your SOC 2?