Infrastructure monitoring
What it costs to self-host Prometheus + Grafana
The Prometheus monitoring system and time series database. Below is the server it needs at each fleet size, what that costs, and how long the 1 commercial product it replaces would take to cost more.
| Servers monitored | Memory | Disk | Cheapest plan that fits | Server | With backups and your time |
|---|---|---|---|---|---|
| 5 servers | 1.3 GB | 11 GB | Starter 2 GB · VPSServer | $10.00 | $124 |
| 10 servers | 1.4 GB | 12 GB | Starter 2 GB · VPSServer | $10.00 | $124 |
| 25 servers | 1.6 GB | 15 GB | Starter 2 GB · VPSServer | $10.00 | $124 |
| 50 servers | 1.9 GB | 20 GB | Starter 2 GB · VPSServer | $10.00 | $124 |
| 100 servers | 2.5 GB | 30 GB | Starter 4 GB · VPSServer | $22.00 | $139 |
| 250 servers | 4.3 GB | 60 GB | Starter 8 GB · VPSServer | $39.00 | $159 |
| 500 servers | 7.3 GB | 110 GB | Starter 8 GB · VPSServer | $39.00 | $159 |
The sizing model for Prometheus + Grafana
Prometheus collects the metrics and Grafana draws the dashboards, with Alertmanager for alerts and node_exporter on each server. Disk follows Prometheus's own capacity formula at its default 15-day retention: about 1,000 series per server scraped every 15 seconds at 2 bytes per sample is roughly 0.2 GB per server. Memory per server is our estimate for the same series count.
- 1 GB of memory for the whole stack at rest, every container the project ships included.
- 10 MB more per monitored server, then 25% on top so the box is not running at its ceiling.
- 10 GB of disk to begin with, growing 0.2 GB per server.
- 2 vCPU. Carried through from the project's baseline and not scaled with fleet size, because for this workload memory is what runs out first.
The base figures are our estimate, derived from the services in the project's reference Docker Compose file, because it does not publish a minimum specification we could cite. The per-server increments are ours in every case. They are deliberately generous rather than optimistic: a comparison that undersizes the server flatters self-hosting.
Time, not just money
Prometheus + Grafana is rated medium to operate, which in this model means about 6 hours to get it running properly the first time and 1.5 hours a month after that. At $75 an hour that is $450 once and $112 a month, which is usually the largest line in the whole calculation.
What it replaces
| Commercial product | Its cost at 25 servers | Prometheus + Grafana at 25 servers | Break-even | |
|---|---|---|---|---|
| Datadog Infrastructure Pro | $375 | $124 | about a month | full breakdown |
Where to put Prometheus + Grafana
1.6 GB of memory and 15 GB of disk is what this needs at 25 servers. The cheapest plan that fits, among the providers whose prices we have read ourselves, is the Starter 2 GB at VPSServer for $10.00 a month.
We are not claiming this is the cheapest server on the internet. It is the cheapest one we can honestly quote. Hetzner Cloud, Vultr and PikaPods publish prices we could not read programmatically, and Hetzner in particular is usually cheaper per gigabyte of memory than anything quoted above. Price them yourself before you commit.
The VPSServer link above is an affiliate link, run through the same programme as our Kamatera link; VPSServer and Kamatera belong to the same group. If you sign up through it we may earn a commission, but the programme does not state an amount for VPSServer. That never changes the computed costs, which server is chosen, or the conclusion of this page. The sizing model and every price source are published in full.
Project facts
| Licence | Apache License 2.0 |
|---|---|
| Licence caveat | Prometheus is Apache-2.0, but Grafana, used here for the dashboards, is AGPL-3.0. That only matters if you modify Grafana and offer it to others over a network. |
| Written in | Go |
| GitHub stars | 66,156 |
| Open issues | 916 |
| Latest release | v3.14.0 |
| Last commit | 2026-09-21 |
| Repository checked | 2026-09-21 |
| Project | prometheus.io · prometheus/prometheus |
Get the migration checklist for this tool
A one-page pre-flight list: data export path, sizing, backup plan, and the rollback step most people skip.