Observability Pricing Models Explained
Per-host, per-GB, per-user, flat-fee. Each pricing model creates different incentives for how you use observability. Understanding these tradeoffs helps you predict costs and avoid surprises.
Vendor pricing comparison (2025-2026)
| Vendor | Primary Model | Data Pricing | User Pricing |
|---|---|---|---|
| Datadog | Per-Host | $0.10/GB logs | N/A |
| New Relic | Per-User + Data | $0.40-0.60/GB | $49-349/user |
| Splunk | Per-GB/Workload | $1,800-2,500/GB/day | N/A |
| Dynatrace | Consumption | $0.20/GB logs | N/A |
| Grafana Cloud | Usage-based | $0.40/GB logs | $8/user |
| Honeycomb | Per-Event | Included | None |
| Sampleless | Flat Annual | Included | Unlimited |
Per-host model analysis (Datadog)
Per-host pricing originated when infrastructure was static. You had servers, you paid per server. Simple.
What counts as a "host"
- Physical servers, VMs, cloud instances: Each OS instance
- Kubernetes: Billed per node, not per pod
- AWS Fargate: Billed based on average task count
The container trap
Deploying the Datadog agent per container/pod instead of per node results in each pod being billed as a host. We have seen teams accidentally turn 50-node clusters into 500-host bills.
The autoscaling penalty
Datadog uses 99th percentile billing. A 5-day traffic spike from 100 to 200 hosts means paying the 200-host rate for the entire month.
Autoscaling for resilience becomes autoscaling your observability bill.
Per-user model analysis (New Relic)
| User Type | Price | Access |
|---|---|---|
| Basic | Free | Dashboards, alerts only |
| Core | $49/user/mo | Logs, telemetry, errors |
| Full Platform (Pro) | $349/user/year | All 50+ capabilities |
The collaboration problem
Per-user pricing discourages giving full access to engineers who need debugging access. Teams often limit access to control costs, creating information silos during incidents.
When the on-call engineer cannot see the same data as the SRE team, incident resolution slows down.
Per-GB model analysis (Splunk)
Splunk charges $1,800-2,500 per GB per day annually at enterprise scale. Volume-based licensing with complex tiers.
The logging dilemma
When every log line costs money, teams face constant pressure to log less. Debug logging gets disabled. Verbose error context gets trimmed. When incidents occur, the evidence is not there.
Pricing model incentive alignment
| Model | Perverse Incentive | Impact |
|---|---|---|
| Per-Host | Use fewer, larger instances | Architecture decisions driven by billing |
| Per-GB | Log less, sample more | Reduced visibility |
| Per-User | Limit access | Collaboration barriers |
| Custom Metrics | Avoid business metrics | Less insight |
| Flat-Fee | None | Full visibility encouraged |
The self-hosted illusion
"We will just run Prometheus and Grafana ourselves."
Self-hosted observability has hidden costs:
- SRE time: 0.5-1.5 FTE for maintenance
- US SRE fully-loaded cost: $250,000-300,000/year per FTE
- Storage and compute: Often underestimated
- Scaling challenges: Prometheus does not scale horizontally without Thanos/Cortex
Self-hosted often exceeds managed service subscriptions while providing less capability and requiring constant engineering attention.
Flat-fee model (Sampleless)
Sampleless uses flat annual pricing based on cloud account count:
- Growth: $100K/year (up to 5 cloud accounts)
- Scale: $180K/year (up to 20 cloud accounts)
- Enterprise: $275K/year (unlimited cloud accounts)
What is included
- Unlimited data volume (traces, logs, metrics)
- Unlimited custom metrics
- Unlimited users
- ALBA behavioral analytics
- Cross-cloud federation
Why flat-fee works
When the vendor does not charge more for using the product more, incentives align. You want to collect everything to maximize visibility. We want you to collect everything because BYOC eliminates our egress costs.
No perverse incentives. No architecture decisions driven by billing. Just observability.
Frequently asked questions
Which observability pricing model is best?
Flat-fee pricing aligns vendor and customer incentives best because there are no perverse incentives to reduce visibility. Per-host, per-GB, and per-user models all create scenarios where using the product more costs more, discouraging full utilization.
Why does Datadog use per-host pricing?
Per-host pricing was common when infrastructure was static. It becomes problematic with autoscaling and containers because costs scale with infrastructure changes, not actual value delivered. Datadog adds per-GB and per-metric charges on top.
Is self-hosted observability cheaper?
Rarely. Self-hosted Prometheus/Grafana requires 0.5-1.5 FTE SREs for maintenance. At $250K-300K/year fully-loaded cost per FTE, this often exceeds managed service subscriptions while providing less capability.