Running your own monitoring stack sounds cheaper than a SaaS subscription until you price it honestly. I run mine on dedicated infrastructure, so here are the real numbers and the parts nobody puts in the spreadsheet.
The compute floor
A minimum viable internal setup needs two machines, not one. One box means your monitoring dies in the same incident as your product, which is the exact moment you need it most. Two small cloud instances in different datacenters run you around 10 to 30 EUR per month combined, depending on the provider and how much you oversize.
That floor covers a single-node uptime checker and maybe a small metrics stack. It does not cover HA, long retention or multi-region probes.
The part that costs real money
Everything after the floor is redundancy. Each of these roughly doubles your bill:
- A second probe location, so a regional network issue cannot fake or hide an outage
- A replicated database, because a monitoring system that loses its own data on failover is a placebo
- A load balancer or failover IP in front, so the public status page stays reachable during a region loss
My own stack (two Hetzner regions, Kubernetes, a stretched database, replicated cache, edge proxies on both sites) lands in the low hundreds of euros per month in raw compute. That is the honest price of monitoring that survives the thing it is monitoring.
The cost that dwarfs the invoice
Engineering time. Building a probe scheduler, check runners for each protocol, alert routing, a status page frontend and an API takes months of evenings before it is reliable. Maintaining it is a standing tax: dependency updates, false positive tuning, storage pruning, on-call for the monitoring itself.
If your hourly rate prices that time at anything realistic, a SaaS subscription pays for itself for the first year or two, easily.
When internal infrastructure wins anyway
- Data residency requirements that no SaaS contract satisfies
- Checking things SaaS cannot reach: private subnets, game protocols, internal services, proprietary stacks
- Scale economics: at hundreds of monitors with aggressive intervals, per-seat SaaS pricing gets absurd while your fixed compute stays flat
- Learning value. I do not regret building mine for a second.
The honest comparison
| Internal infra | SaaS | |
|---|---|---|
| Cash cost | 30-300+ EUR/mo compute | 0-100+ EUR/mo per plan |
| Time cost | Weeks to build, hours per month to keep | Near zero |
| Multi-region | Pay per site you add | Usually included |
| Data control | Full | Contractual |
| Failure blast radius | Yours to engineer away | Their problem |
Internal monitoring infrastructure is not cheap. It is expensive in a currency most estimates ignore: your weekends. If you have the data or scale reasons to justify it, the cash side is fine. If you do not, buy the subscription and spend the months on your product instead.