Managed vs. Self-Hosted Observability: The Real Cost Comparison

The self-hosted vs managed debate isn't about license cost — it's about total cost including engineering time, reliability risk, and opportunity cost.

The True Cost Model

Managed (SaaS) Costs

License fees + agent overhead. That's it. No infra, no ops, no upgrades.

Self-Hosted Costs

1. Infrastructure: Compute, storage, network for the observability stack itself
2. Engineering time: 1-3 FTEs for setup, maintenance, upgrades (fully loaded: $150-300K each)
3. Reliability risk: Self-hosted outage during a production incident = double crisis
4. Feature development: Building integrations, dashboards, alerting that vendors provide out of box
5. Opportunity cost: Those engineers could be building product features

Break-Even Analysis

Scale	SaaS Monthly	Self-Hosted Monthly	Break-Even?
Small (20 hosts)	$2-5K	$5-8K (infra + 0.25 FTE)	No
Mid (200 hosts)	$15-40K	$10-20K (infra + 1 FTE)	Maybe
Large (1,000 hosts)	$50-150K	$15-40K (infra + 2 FTE)	Yes
Enterprise (5,000+ hosts)	$200-500K	$30-80K (infra + 3 FTE)	Definitely

Self-hosted typically breaks even around $30-50K/month SaaS spend (200-500 host range).

The LGTM Stack (Self-Hosted)

Loki (logs) + Grafana (visualization) + Tempo (traces) + Mimir (metrics). All open-source, all OTel-native.

Realistic infrastructure for 200 hosts:

• 3 Mimir nodes (metrics): ~$600/month
• 3 Loki nodes (logs): ~$800/month
• 1 Tempo node (traces): ~$200/month
• 1 Grafana instance: ~$100/month
• S3 storage: ~$200/month
• Total infra: ~$1,900/month
• Total with 1 FTE: ~$14,400/month

Compare to $15-40K/month SaaS.

When Self-Hosted Wins

• SaaS spend exceeds $30K/month
• You have strong platform engineering team
• Data sovereignty requirements (GDPR, regulated industries)
• Need customization vendors don't support
• Already invested in Prometheus/Grafana ecosystem

When Managed Wins

• SaaS spend under $20K/month
• Small engineering team (<50 engineers)
• Need broad integration ecosystem out of box
• Observability is not a core competency
• Speed to value matters more than cost optimization

The Hybrid Approach

Many organizations land on hybrid: self-hosted for high-volume, well-understood workloads (infrastructure metrics, application logs) and SaaS for specialized capabilities (APM, synthetics, RUM). This captures 60-70% of self-hosted savings without giving up vendor capabilities entirely.

Break-even sketch (500-host estate)

Managed (Datadog-class)	Self-hosted LGTM on AWS
Platform $	~$45K/mo	~$8K infra + $25K SRE (0.5 FTE)
Ops burden	Low	High (upgrades, scaling, on-call)
Time to value	Days	3–6 months hardening

Break-even often appears at 800–1,500 hosts or >500 GB/day logs — if you already have platform SRE capacity. Otherwise managed wins on total cost of ownership.

What to do this week

• [ ] Model self-hosted infra with calculator LGTM preset
• [ ] Add 0.5–1 FTE fully loaded cost to self-hosted column
• [ ] List features you'd lose (RUM, synthetics, ML alerts)
• [ ] Hybrid: managed for APM, self-hosted for logs/metrics?

Sources & further reading

• OpenTelemetry cost guide
• Grafana LGTM stack docs — operational requirements

---

Related Reading

• Comparing TCO
• OpenTelemetry Cost Guide
• Evaluating Datadog Alternatives
• Reducing Infrastructure Monitoring Costs

Use the SignalCost Calculator → to model these scenarios with your own numbers.