Datadog pricing: what the bill is made of, and five alternatives

What the Datadog bill is made of

Datadog sells per product, and every line is a count multiplied by a rate. The pricing page lists the units: infrastructure hosts, APM hosts, indexed log events, indexed spans, RUM sessions and custom metrics. A host running the Agent pays once; instrument it for tracing and it pays again.

Logs show the two-part shape. You pay to send data in, and again for the events kept queryable in an index, so what arrives and what stays searchable are separate numbers. Datadog totals indexed events monthly: above the commitment, on-demand usage carries a 50% premium.

The annual commit frames the rest. List rates are per host or per unit of volume; a negotiated agreement substitutes committed volumes and on-demand rates above them. That is where the discount lives and where renewal starts.

The two lines that grow without a decision

A custom metric is not a metric name. It is a name plus its tag values, host tag included, and the billable figure is the monthly average of distinct combinations per hour. Pro includes 100 indexed custom metrics per host, Enterprise 200, counted across the estate; the rate above that sits in the contract.

On a 200-host Pro estate the allotment is 20,000 indexed custom metrics, so one noisy service emitting 40,000 tag combinations can consume the whole account's allowance twice over.

Instrumentation moves it. A histogram or distribution counts as five custom metrics per unique name-and-tag combination, and percentile aggregations double that, so a busy service can go from four combinations to forty without anyone approving a purchase. Metrics without Limits allowlists queryable tags and charges $0.10 per 100 ingested custom metrics above the allotment.

Log indexing grows the same way. A new debug level, a chatty sidecar or a Kubernetes event stream changes index volume, and the bill follows events kept rather than events received. One bad day rarely moves a monthly number; a pipeline that indexes everything moves it monthly.

The commit, the true-up and the renewal call

An agreement is usually a committed volume per product, with rates inside the commitment and on-demand rates above it. Log commitments are monthly; custom metrics and indexed spans behave the same way, with the overage price written into the contract.

Bring four numbers to the renewal: twelve months of usage by product line, the committed volume for each, the on-demand rate above it, and the tag configurations on your largest custom metrics.

Discounts follow volume you can prove and volume you can remove. Consolidation onto one vendor is worth arguing about, and so is a plan to cut cardinality by tag configuration, because the vendor reads the same usage page.

Grafana Cloud and New Relic: the two usage-based options

Grafana Cloud is the closest thing to a like-for-like replacement. It is assembled from open-source components many teams already run, and prices usage against published allowances rather than per host: metrics by series, logs and traces by gigabyte. Its free tier runs a small estate at no cost.

New Relic prices on two axes: seats on the full platform and data ingested. The free tier publishes a monthly ingest allowance and a full-platform seat, and metrics, logs and traces draw on the same allowance, so the bill is easy to forecast and easy to grow. A verbose log line costs what a metric costs.

For a team fluent in PromQL and Grafana dashboards, Grafana Cloud shortens the migration because the query language is already theirs. For one that wants seat-based governance and a single ingest number, New Relic is shorter.

Elastic, CloudWatch and the self-hosted floor

Elastic Observability fits an estate already running Elasticsearch for search, security analytics or log storage, because the logs stay where they are and the query language is in use. Which contract is on the table matters: the serverless model and the self-managed licence put operational work on opposite sides.

CloudWatch needs no procurement. Custom metrics are billed per metric per month in tiers, logs per gigabyte ingested and stored, queries by volume scanned, all on the AWS invoice with an always-free allowance. The ceiling is scope: infrastructure and logs with tracing in X-Ray, not a service catalogue with error budgets.

The self-hosted floor is Prometheus, Loki and Tempo behind an OpenTelemetry Collector that keeps the exporter a configuration decision rather than a rewrite. No licence line, but a storage bill, retention and high-availability decisions, and a person who owns them. A single node is a weekend.

What switching actually costs

Dashboards are the cost nobody prices. A dashboard is a query, a layout and a set of visual conventions, and only the query ports cleanly: a PromQL panel becomes a Grafana panel with a rewrite.

Monitors carry more than a threshold. A working alert encodes which metric is trustworthy, what normal looks like at that hour, who gets paged and what the responder does first. Thresholds and runbook links export; the reasoning sits with the engineers who built them. Run both systems in parallel.

Compare three numbers: the annual licence and commit, the engineer weeks to rebuild the top twenty dashboards and fifty monitors, and the storage and egress for anything self-hosted. Halving the bill while doubling the on-call load is not a saving.

Choose this if

Stay on Datadog if the platform replaces three tools at once, with APM, logs, RUM and security signals on one query language, and the estate is stable enough to manage cardinality by tag configuration.

Move to Grafana Cloud if the team lives in PromQL; to New Relic if seat-based governance suits; to Elastic if the logs already sit in Elasticsearch; to CloudWatch if workloads are almost entirely on AWS; to Prometheus, Loki and Tempo only if a platform team wants to own storage.

Then hold the renewal with the usage page open. The two lines that grew without a decision can shrink without one: tag configuration on custom metrics, and an index rule for logs.