A team running 20 hosts with moderate logging and APM enabled checked their Datadog invoice and found it had quietly reached $3,000-8,000/month - not because they'd added infrastructure, but because Kubernetes' high-cardinality labels and short-lived pods had pushed custom metrics and log indexing well past what the advertised per-host price implied. Datadog's $15/host sticker price was real. It was also a small fraction of the actual bill.
That gap between advertised pricing and real Kubernetes-scale cost is the central question in choosing between Prometheus, Datadog, and New Relic. All three can monitor a Kubernetes cluster. They differ enormously in what triggers the bill to grow, and in how much operational ownership you're signing up for in exchange for a lower number.
The three tools solve observability from opposite ends: one is free software you operate yourself, two are managed platforms billed on usage.
Prometheus: Open-source, self-hosted, zero licensing cost.
Pull-based metrics, [PromQL](/glossary/promql), pairs with [Grafana](/glossary/grafana)
for dashboards. Requires platform engineering
capacity for HA, storage, and alerting upkeep.
Datadog: SaaS, priced per product - infrastructure
($15/host), APM ($31/host), logs ($0.10/GB +
indexing), custom metrics beyond allowance.
900+ integrations, broadest feature surface.
New Relic: SaaS, priced per full-platform user plus data
ingest ($0.30/GB over 100GB free). Free tier:
1 user, 100GB/month. Increasingly competitive
at scale, especially for APM-heavy workloads.
Prometheus's entire trade is operational ownership for zero licensing cost - you get full control and no vendor lock-in, in exchange for owning every part of running the stack yourself, from storage retention to alerting rule maintenance. Datadog's trade is breadth and polish at a price that compounds quickly on Kubernetes specifically, because its per-product, per-host pricing model wasn't designed around the high-cardinality, ephemeral nature of container workloads. New Relic's trade is a genuinely useful free tier and increasingly competitive scaled pricing, with the caveat that its per-user fee still ties total cost to headcount in a way a self-hosted tool never does.
For a representative scenario - 10 Kubernetes nodes, 5 microservices, moderate log volume around 50GB/month, distributed tracing enabled - Prometheus wins outright on raw cost, since the only expense is the compute it runs on. That comparison assumes existing platform engineering capacity absorbs the operational work; it is not free in the sense of requiring zero people, only in the sense of no licensing fee.
New Relic's 2026 pricing shift makes it surprisingly competitive at this scale and larger. The free tier - one full-platform user, 100GB/month of ingest - is genuinely usable for small teams, not a crippled trial. Paid tiers start around $49/month per additional full user plus $0.30/GB over the free allowance, and at larger team sizes this is generally cheaper than Datadog, particularly when APM is the primary need over deep infrastructure monitoring.
Datadog's real cost at this same scale routinely lands at $3,000-8,000/month for teams running 20 hosts with moderate logging and APM enabled - not because any single line item is unreasonable, but because infrastructure, APM, logs, and custom metrics are billed as four separate products that all scale up simultaneously with Kubernetes-driven cardinality.
kubectl get pods --all-namespaces | wc -l
Kubernetes creates exactly the traffic pattern that per-host, per-metric pricing models weren't built for: high-cardinality labels, short-lived pods that come and go far faster than a traditional VM host, noisy logs, many distributed tracing spans, and custom metrics that multiply with every microservice added. Datadog charges separately for custom metrics beyond the per-host allowance at $5 per 100 metrics, and separately again for log indexing on top of the $0.10/GB ingestion charge - a Kubernetes cluster emitting thousands of high-cardinality labels can hit both charges simultaneously without anyone noticing until the invoice arrives.
Datadog's APM product includes 150GB of trace spans per host before overage charges kick in; high-throughput microservices architectures routinely blow past that allowance quietly, since span volume scales with request volume and service count, not with host count. None of these overage mechanisms are unique to Datadog conceptually - New Relic's per-GB ingest model has the same shape - but Datadog's per-product billing structure means Kubernetes cardinality growth compounds across more separate line items at once.
Pricing Model and Operational Cost:
| Factor |
Prometheus |
Datadog |
| Licensing cost |
$0 - self-hosted |
~$15/host infra, ~$31/host APM |
| Operational ownership |
High - HA, storage, alerting all self-managed |
Low - fully managed SaaS |
| Kubernetes cardinality risk |
Storage cost scales with retention choices |
Custom metrics + log indexing overages |
Pricing Model and Operational Cost, continued:
| Factor |
New Relic |
Best fit |
| Licensing cost |
Free tier: 1 user/100GB; then per-user + $0.30/GB |
Teams needing predictable, moderate scale cost |
| Operational ownership |
Low - fully managed SaaS |
Teams without dedicated platform engineering |
| Kubernetes cardinality risk |
Ingest cost scales with log/trace volume |
APM-primary workloads over deep infra monitoring |
Default to Prometheus and Grafana when your team already has platform engineering capacity and wants to avoid per-host or per-GB billing tied directly to Kubernetes' natural cardinality growth - the cost predictability is real, but only if someone genuinely owns the operational upkeep, not as an unstaffed side responsibility.
Choose Datadog when breadth matters more than cost predictability - security monitoring, RUM, service management, and 900+ integrations in one platform genuinely reduce tool sprawl for enterprises running many environments. Budget explicitly for custom metrics and log indexing as separate line items from day one rather than discovering them at invoice time, since Kubernetes-driven cardinality growth is the most common cause of Datadog bill surprises.
Choose New Relic when APM is the primary need and team size is moderate and predictable - the free tier is a legitimate starting point, not just a trial, and 2026's per-user-plus-ingest pricing tends to undercut Datadog at scale. For Indian engineering teams optimizing observability spend against a growing Kubernetes footprint, running a real cost comparison against your specific node count, log volume, and team size before committing matters more than any general ranking between the three.
INFORMATIONReferences and Further Reading
Discussion0