Observability vendors split into roughly two pricing philosophies, and picking the wrong one for your architecture is the most common reason bills surprise people. Here is how they differ in 2026.
How the two models work
| Model | You pay for | Vendors | Best for |
|---|---|---|---|
| Per host | Each monitored host/node | Datadog, Splunk, Dynatrace | Stable fleet of large hosts |
| Per vCPU | Each monitored core | AppDynamics | Compute-sized billing |
| Per GB / usage | Telemetry data volume | Grafana, Elastic, Coralogix, New Relic | Dense Kubernetes, ephemeral hosts |
| Per event | Events/spans/errors | Sentry, Honeycomb | Trace/error-first, low volume |
The Kubernetes problem with per-host
Per-host pricing assumes hosts are few and long-lived. Modern Kubernetes breaks that assumption: you might run dozens of small pods across many nodes, or autoscale hosts up and down by the minute. With per-host billing you pay for every node regardless of how little it emits, and ephemeral hosts can each count. This is why dense-Kubernetes teams often move to usage or per-vCPU models. Compare Datadog vs Grafana Cloud to see the divide in numbers.
The verbose-telemetry problem with usage-based
Usage-based pricing has the opposite failure mode. If your apps log everything at debug level, emit high-cardinality metrics, or trace every request without sampling, your per-GB or per-event bill balloons even on a handful of hosts. Usage models reward disciplined instrumentation and punish noise. See how to cut observability costs for the levers.
A simple decision rule
- Few, large, stable hosts + lots of telemetry -> per-host can be cheaper and more predictable.
- Many small / ephemeral hosts (Kubernetes) + controlled telemetry -> usage-based usually wins.
- Trace/error-first workload at low volume -> per-event (Sentry, Honeycomb) is often cheapest.
Bottom line
There is no universally cheaper model - only the right model for your scaling shape. Estimate both in the observability cost calculator and browse the full vendor list. Prices are a June-2026 snapshot; verify on each vendor’s pricing page.