Observability budgets surprise engineering teams because costs scale on multiple independent axes simultaneously — hosts, containers, log volume, APM services, custom metrics, and synthetic monitors. A team that estimates $3,000/month on Datadog often lands at $15,000 once all dimensions are counted.
Datadog pricing structure: Infrastructure monitoring at $15-23/host/month (Standard/Pro). APM at $31/APM host/month (billed per host running traced services, separate from infra). Logs at $0.10/GB ingested + $0.05/GB indexed (retention-based). Custom metrics at $0.05 per metric/month (first 100 free per host). Containers in Kubernetes add $1/container/month with a minimum of 10 containers. A 50-service deployment on 30 hosts with 150 containers, 500GB/month logs, and 2,000 custom metrics costs approximately $8,000-12,000/month.
New Relic's pricing model changed in 2022 to user + data model. $0.35/GB after the first 100GB free, plus $99/month per full-platform user. For data-heavy teams, this can be significantly cheaper than Datadog. A team of 10 engineers ingesting 1TB/month pays roughly $10 × $99 + (900GB × $0.35) = $990 + $315 = $1,305/month — a dramatic difference.
Honeycomb targets high-cardinality event data. Pricing is per event at $130/million events beyond the free tier (20M events/month). For microservices with high request volume, this adds up quickly — 1 billion events/month costs roughly $130,000. Honeycomb makes sense when you need to query against high-cardinality fields (user IDs, request IDs, custom business attributes) that Datadog and Grafana index poorly.
Grafana Cloud (self-hosted alternative): $0 for the first 50GB logs, 10K metric series, and 50GB traces. Beyond that, approximately $700-2,000/month for a medium engineering org. The tradeoff is operational overhead if self-hosting Grafana OSS stack (Loki, Mimir, Tempo) versus the managed cloud service.
Practical cost reduction: sample traces at 10-30% for non-error paths (you rarely need 100% trace fidelity), aggregate low-cardinality metrics client-side before shipping, set retention policies aggressively (hot storage for 7 days, cold for 30), and filter DEBUG logs at the collector before ingestion.