The cost and architecture of Grafana enterprise observability

Datadog charges $15 per host per month for infrastructure monitoring and $31 per host per month for APM on annual plans. This per-host model punishes container-dense fleets because every sidecar and node increases the bill. A team running containerized workloads or anything with high-cardinality metrics can watch their bill climb in ways that have little to do with how much value they get out of the tool. Datadog also auto-generates a large volume of custom metrics from OpenTelemetry instrumentation, which often leads to unexpected invoices when teams realize they are paying for thousands of metrics they did not explicitly configure.

Grafana Cloud bills metrics at $6.50 per 1,000 active series on its Pro tier. If a user sets a scrape interval to 15 seconds, the billing switches to total data points per minute, which can quadruple the metrics bill compared to standard scraping. New Relic uses a different model, charging per user and for data ingest of $0.35 per GB after the first 100 GB. This makes New Relic attractive for high-cardinality workloads where the metric count does not directly drive cost.

Scaling Prometheus and the operational load

Teams prioritizing cost control and total ownership often choose the open-source path using Prometheus for metrics and Grafana for visualization. This stack costs essentially what the hosting and engineering time cost, as there are no per-metric fees. For a 10M time series deployment, this remains dramatically cheaper than any SaaS alternative. You can use Thanos for long-term storage to achieve the best price-to-performance ratio, whereas Mimir via Grafana Cloud reduces operational overhead but costs 2 to 4 times more than DIY Thanos.

The tradeoff involves heavy operational responsibility. If your platform team has fewer than four engineers, you should skip self-hosted Grafana because running Loki, Mimir, and Tempo at scale imposes a massive load. You must manage uptime, scale the storage backend as retention grows, and tune query performance as cardinality climbs. If you choose to manage the stack yourself, you inherit the on-call burden for the observability infrastructure.

If you use OpenTelemetry to instrument your services, you avoid vendor lock-in because switching backends requires only a collector configuration change. For teams that want the same visualization ecosystem and query flexibility without owning the operational burden, Grafana Cloud provides a middle path between full self-hosted control and a fully managed SaaS platform.

Architecture for metrics and business data

Grafana acts as a visualization layer that sends queries to existing data sources and renders the results. It excels at real-time operational dashboards for API latency, error rates, and container health. Because it supports sub-second refresh intervals and works natively with Prometheus via PromQL, it remains the industry standard for engineering teams.

Engineering teams often struggle when they try to repurpose Grafana for customer-facing business reporting. Grafana lacks a native SQL database query builder for non-technical users, meaning every panel requires a developer to write and maintain raw SQL. It also lacks a concept of paginated reports for PDF or print delivery, so business users cannot easily generate branded invoices or compliance exports.

New Relic offers a different approach by using NRQL, a SQL-like language that simplifies the learning curve for teams coming from a database background. New Relic also provides automated root-cause analysis via its Davis AI engine, which correlates signals across the stack to find causal chains. Datadog provides the fastest path to production dashboards through its massive integration catalog and unified platform for metrics, APM, and logs.

The decision between Datadog, New Relic, and Grafana depends on your scale. If you run a container-dense fleet on a tight budget, skip Datadog. If you have many users who only need to glance at dashboards, skip New Relic because per-user pricing punishes large organizations. If your team lacks the engineering headcount to manage a custom stack, skip self-hosted Prometheus. How will your team manage the sudden spike in metric cardinality during your next deployment?