Luca BertonObservability Cost Is the Elephant in the Room
Every enterprise running Kubernetes at scale faces the same question: how much are we spending on observability, and is it worth it? Datadog's per-host and per-GB pricing model makes costs predictable but expensive at scale. The Grafana stack (Prometheus, Loki, Tempo, Mimir) is open-source but requires engineering investment to operate.
For regulated enterprises, there's a third dimension: data residency and compliance. Where does your telemetry data live, and who can access it?
Head-to-Head Comparison
| Dimension | Datadog | Grafana Stack |
|---|---|---|
| Model | SaaS (managed) | Self-hosted or Grafana Cloud |
| Metrics | Custom metrics (per metric) | Prometheus/Mimir (unlimited self-hosted) |
| Logs | Log Management (per GB ingested) | Loki (per GB or free self-hosted) |
| Traces | APM (per host + per span) | Tempo (free self-hosted, per GB on Cloud) |
| Dashboards | Excellent built-in | Grafana (industry-leading) |
| K8s monitoring | Datadog Agent (comprehensive) | kube-prometheus-stack |
| AI/ML monitoring | LLM Observability (2024) | Custom dashboards + exporters |
| Alerting | Monitors (ML-powered anomaly) | Grafana Alerting + Alertmanager |
Cost Comparison at Scale
100 Kubernetes Nodes, Full Observability
- Datadog: ~$200K-400K/year (infrastructure + APM + logs + custom metrics). Cost scales with every new node, container, and custom metric.
- Grafana Cloud: ~$50K-150K/year (depends on metrics and log volume). More predictable cost model based on data volume.
- Self-hosted Grafana Stack: ~$30K-80K/year (infrastructure cost) + 1-2 FTE to operate = ~$180K-380K total. Similar TCO to Datadog but with full data control.
The cost trap: Datadog costs often surprise enterprises at renewal. Each new team adding custom metrics or log pipelines increases the bill. Budget for 30-50% year-over-year growth in Datadog costs.
Kubernetes Recipes
A practical guide for container orchestration and deployment by Grzegorz Stencel & Luca Berton (Apress).
Watch on Skillshare →When Datadog Wins
- Time to value — Agent install → full observability in hours. No infrastructure to manage.
- Correlation — Seamless metrics ↔ traces ↔ logs ↔ infrastructure correlation in one platform.
- AI features — Watchdog anomaly detection, ML-powered alerting, LLM Observability are genuinely useful.
- Small-medium scale — Under 50 nodes, Datadog's cost is reasonable and the operational savings are real.
When Grafana Stack Wins
- Data sovereignty — Self-hosted means telemetry data stays in your infrastructure. Critical for GDPR and NIS2.
- Cost at scale — Beyond 100 nodes, self-hosted Grafana stack costs a fraction of Datadog.
- Customisation — Prometheus exporters for anything. Grafana dashboards are the industry standard for a reason.
- No vendor lock-in — OpenTelemetry + Prometheus + Grafana = fully portable observability stack.
- DORA compliance — Self-hosted observability data = full control over retention, access, and audit trails for ICT risk management (Art. 5-16).
Microsoft SQL Server Performance Tuning
Performance tuning essentials for SQL Server. In collaboration with Starweaver.
Start on Coursera →Hybrid Approach
Many regulated enterprises use both:
- Grafana Stack self-hosted for infrastructure metrics, compliance logs, and security events (data residency)
- Datadog for APM and developer experience (traces, error tracking, RUM)
This gives data sovereignty where it matters and developer productivity where it counts.
Decision Framework
Choose Datadog if: Under 50 nodes, no data residency requirements, value time-to-value over cost optimisation.
Choose Grafana Stack if: 100+ nodes, data sovereignty requirements, engineering team to operate it, cost-sensitive at scale.
Consider hybrid if: Need data sovereignty for compliance data but want SaaS developer experience for APM.
AI Platform Assessment
Get a 2-3 week infrastructure audit with a concrete roadmap. No big-consultancy overhead.
Book Your Free Assessment →