Datadog → Grafana + Prometheus
What it saves, what actually moves, what you rebuild — and the thing that catches people.
What it costs, and what it saves
Datadog bills roughly $15–23 per host per month for infrastructure plus $0.10 per ingested GB of logs, and real bills land between $200 and $2,400 a month. Grafana plus Prometheus on a $20 VPS costs the VPS. A 20-host estate typically drops from about $600/month to under $50.
Moves cleanly
Nothing, and that is the honest headline. Datadog holds your historical metrics and there is no export that Prometheus can ingest. You start a new time series on day one.
You rebuild
Every dashboard, every monitor, every alert route. Datadog's query language is not PromQL and there is no converter worth trusting.
What Datadog costs you today
Per-host and per-feature billing that scales quickly: infrastructure monitoring starts around $15–23 per host/month, APM and log management are billed on top (logs per-GB ingested and per-million events retained), and custom metrics add further cost — large fleets routinely see five- and six-figure annual bills.
What actually holds you in
Proprietary agents, proprietary query languages, and per-host/per-feature pricing mean both your cost and your data gravity grow with your fleet — the more you instrument, the harder and more expensive it is to leave.
What you are moving to
Prometheus scrapes and stores metrics; Grafana visualizes them (and logs and traces via Loki and Tempo). Together they are the de-facto open-source monitoring stack, fully self-hostable.
Free / self-host
Grafana + Prometheus strengths
- Industry-standard, huge ecosystem
- Metrics, logs, and traces via the same UI
- Enormous community and dashboards
What you give up
- You assemble and operate the stack
- Steeper initial setup than a SaaS agent
The migration, step by step
- 1Stand up Prometheus and Grafana — docker compose with the kube-prometheus-stack chart if you are on Kubernetes
- 2Install node_exporter on each host and point Prometheus at it
- 3Run both systems in parallel for one full billing cycle so you can compare graphs side by side
- 4Rebuild your five most-used dashboards only — you will find you never rebuild the rest
- 5Move alerting to Alertmanager, test every route with a deliberate failure, then cancel Datadog
The gotcha
Prometheus is not long-term storage. Default retention is 15 days and it stores locally. If you need a year of history you also need Thanos, Mimir or Cortex, and that is a second system to operate — budget for it or you will discover it three months in.
When to stay on Datadog
You rely on APM tracing, Watchdog anomaly detection, or the security monitoring — those have no clean open equivalent and rebuilding them costs more than the licence.
Other ways off Datadog
More Layer 3 migrations
Figures verified 2026-07-12 against vendor pricing pages. Prices change and migrations differ by estate — treat the cost delta as a starting model, not a quote. Rankings and recommendations here are merit-only; affiliate income never changes a verdict. See our methodology.