APM pricing comparison 2026

APM products span more pricing model variation than any other category in cloud monitoring. The reason is that distributed tracing, the modern core of APM, is fundamentally expensive to capture in full and economically painful to bill in any single dimension. Per-host pricing rewards homogeneous fleets but punishes microservices density. Per-GB ingest pricing scales with telemetry volume but obscures cost-per-trace. Per-agent pricing rewards monolithic applications but punishes Kubernetes pod density. Per-DPS-unit pricing offers consumption flexibility but obscures predictability. Each model trades one set of customer-friendly properties for another.

Datadog APM at $35 per host per month on the Pro plan is the most widely-deployed pricing model, with a clear cost basis (per host, like infrastructure monitoring) and a predictable behaviour as workload scales. The hidden meter is indexed-span overage; Datadog includes 1 million indexed spans per host per month, and high-throughput services routinely exceed this with another $1.70 per million event charge. For services receiving 100 plus requests per second per host, the indexed-span overage can match or exceed the per-host APM line item.

New Relic APM is structurally cheaper for most workloads because the unified ingest meter at $0.40 per gigabyte absorbs APM telemetry without a separate per-host APM charge. A typical APM-instrumented host produces around 0.3 gigabytes per day of trace and metric telemetry, or 9 gigabytes per month. At 100 hosts that is 900 gigabytes per month, or $320 above the 100 gigabytes free; less than 10 hosts of Datadog APM at the same scale. The trade-off is that ingest volume is harder to forecast than host count, and a single noisy application can substantially shift the bill.

Grafana Tempo at $0.45 per gigabyte of trace ingest is the open-standard option, OpenTelemetry-native and architecturally aligned with the broader Grafana Cloud product line. The pricing is transparent and the Cloud TCO is competitive with New Relic at most scales. The trade-off is that Tempo focuses on tracing; teams want APM-style code profiling and language-runtime metrics typically pair Tempo with Pyroscope (continuous profiling, also Grafana Cloud) and Mimir (metrics) to reach Datadog APM-equivalent capability.

Dynatrace at $58 per host (memory-banded) bundles APM into Full-Stack Monitoring alongside infrastructure and real-user monitoring. The headline rate is high but the bundle includes capabilities that Datadog charges separately for, and Dynatrace is explicit that Full-Stack already contains everything in Infrastructure Monitoring rather than stacking on top of it. On a like-for-like full-stack comparison, Dynatrace and Datadog land within 10 to 20 percent of each other at list. Splunk AppDynamics Premium is listed at $33 per host, positioned similarly to Datadog APM, but the licence meters per CPU Core, so a dense multi-core host can cost several times the headline; Cisco channel discounts of 25 to 50 percent below list are routine. Splunk Observability Cloud reaches APM at the App & Infra edition from $60 per host, which likewise includes infrastructure monitoring, though feature velocity trails the leaders.

The single most consequential APM cost-management practice is trace sampling. Most APM workloads default to capturing 100 percent of traces, which produces dramatically more data than is operationally useful and proportionally inflates the APM bill. Sampling at 5 to 10 percent preserves error-rate accuracy and percentile latency for operational analysis while reducing trace volume by 90 percent.

Two sampling strategies dominate. Head-based sampling decides at the start of a trace (when the first span is created) whether to capture the whole trace or drop it. The decision is fast and cheap but can miss interesting traces (an error that occurs deep in the trace path is dropped if the head was unlucky enough to sample at that low rate). Tail-based sampling captures all traces initially, holds them in memory or short-term buffer, and decides at trace completion whether to retain or drop based on outcomes (errors, slow latency, specific spans). Tail-based sampling is operationally superior but requires more sampling infrastructure (the OpenTelemetry Collector or equivalent buffer).

For workloads at meaningful scale, tail-based sampling at the OpenTelemetry Collector level is the most efficient pattern. Capture 100 percent of traces at the application, route them to an OTel Collector, apply tail-based sampling rules (always retain traces with errors, always retain traces with latency above the 99th percentile, retain 5 percent of normal traces randomly), then forward the retained traces to the APM vendor. This typically retains less than 10 percent of traces by volume while preserving more than 95 percent of the operationally important traces.

The economic impact at scale is substantial. A 100-host deployment running unsampled APM at Datadog default settings can produce 10 to 50 million indexed spans per host per month (well above the 1 million included), with overage of $15,000 to $80,000 per month. Tail-based sampling at 10 percent typically eliminates the overage entirely, saving the same $15,000 to $80,000 per month with no operational compromise.