Datadog pricing, explained: how to cut the bill, or what to run on AWS instead.

The biggest saving is usually not switching vendors. It's the gap between the first two rows and the third: stop indexing every log line, stop tagging metrics with things like user IDs, and stop tracing hosts that don't serve requests. That takes days, not a migration.

Switching gets you from roughly $470 to somewhere between $210 and $375 on these numbers, and costs you engineer time to move alerts, dashboards and instrumentation. Whether that's worth it depends on who will own the new setup. Section 12 is our honest take on when Datadog is the right tool.

What each host includes on Pro: 100 custom metrics, 5 containers and 500 custom events. Enterprise includes 200 custom metrics and 10 containers. An APM host includes 150 GB of ingested spans and 1 million indexed spans a month. The allotments are pooled across the account, so 10 Pro hosts give you 1,000 custom metrics to share.

The main pricing page shows annual prices as "starting at", and most companies past the first few months sign a contract with its own rates. We can only price what's public, so treat every Datadog figure on this page as list price.

On the standard high water mark plan, Datadog counts your hosts every hour. At the end of the month it drops the busiest 1% of hours and bills the highest count that's left. Datadog's APM billing page spells it out: the ninth-highest hourly count is billed (the eighth in February), so the top 8 hours are ignored. Anything that lasts longer than that sets your bill for the whole month.

A daily traffic peak, a nightly batch job on extra instances or a blue/green deploy that doubles the fleet for an hour each day can all lift you to the peak count. Datadog also offers a hybrid plan where you commit to a monthly minimum and pay hourly above it, which suits spiky fleets better. Ask for it.

Other counting rules that catch teams out:

• The AWS integration bills EC2 instances without an agent. Datadog bills hosts running the Agent and all EC2 instances the AWS integration picks up. CI runners, bastions and dev boxes count. RDS, ELB and DynamoDB are not billed as hosts.
• Containers are pooled, then charged by the hour. Datadog samples containers every five minutes, averages them per hour and subtracts your allotment. The Datadog Agent's own containers and Kubernetes pause containers don't count. A crash-looping container still counts in any five-minute interval where it runs for more than ten seconds.
• Fargate is billed per task, measured every five minutes, not per host. APM on Fargate is $6 per concurrent task a month on annual billing, the serverless application line above.
• Custom metrics are a monthly average of hourly unique counts. A short burst of new tag values for a few hours raises the average less than you'd fear, but a permanent high-cardinality tag raises it for good.

Index fewer logs, cut high-cardinality tags, sample traces, stop paying for hosts you don't watch, then commit to annual rates. On the example in section 05 that takes the bill from $1,100 to $468 a month.

In the order we'd do them. The first three cover most of the saving on a typical startup bill.

1. Add log exclusion filters. In each log index, exclude health checks, debug level and successful access logs, or sample them (an exclusion filter can keep anywhere from 0% to 100% of matching logs). Excluded logs still show in Live Tail, can still be archived and can still feed log-based metrics, so create a metric for anything you count before you exclude it. You still pay the $0.10 per GB to ingest them; you stop paying to index them.
2. Set a daily quota on each index, so one noisy deploy can't index a month's budget in a day. After the quota, logs stop being indexed but stay in Live Tail, archives and metrics.
3. Pick shorter retention where you can. 7-day indexing is $1.27 per million events against $1.70 for 15 days. Most debugging happens within a week. Archive everything to your own S3 bucket and rehydrate when you need it ($0.10 per compressed GB scanned).
4. Find and cut high-cardinality tags. Look for tags such as user, customer, request, session or container IDs on custom metrics. Remove them at the source, or use Metrics without Limits to keep only the tags your dashboards and monitors query. Check the estimate before saving: Datadog's docs say not to save a tag configuration that shows more indexed than ingested metrics. Turn percentiles off on distributions nobody graphs.
5. Sample traces deliberately. By default the Agent targets 10 traces per second per Agent (DD_APM_TARGET_TPS). Lowering that target is the simplest cut, and the Agent's error sampler still keeps traces with errors. Where you need finer control, DD_TRACE_SAMPLING_RULES keeps 100% of the endpoints that matter (checkout, sign-in) and a low rate for the rest, with a default limit of 100 traces per second per service instance. Datadog's docs note the error and rare samplers are ignored for services with sampling rules. Then remove the tracer from hosts that don't serve requests, which removes the APM host fee.
6. Stop paying for hosts you don't watch. In the AWS integration, limit metric collection by tag, for example !datadog:no, and tag CI runners, bastions and dev instances. Fewer, larger instances also mean fewer billable hosts, since the price is per host, not per CPU.
7. Commit once usage is stable. Moving from on-demand to annual saved $335 a month in section 05 with no change in usage. Do it after the trimming, not before, or you'll commit to the waste. Set alerts on Datadog's estimated usage metrics so the next spike shows up within a day, not on the invoice.

One trap to know before you move: classic CloudWatch custom metrics cost $0.30 each, six times Datadog's $0.05. Sending Prometheus-style metrics there one series at a time can cost more than Datadog did. Send them to Managed Prometheus, or as OpenTelemetry metrics priced per GB.

Log class is chosen when a log group is created and can't be changed later. Put audit, debug and access logs you rarely read into Essentials (formerly Infrequent Access) log groups from the start; keep Standard for logs that drive alarms.

With Transaction Search on, Application Signals is billed per GB of spans, so the APM line scales with traffic, not hosts: double the requests and it roughly doubles. Datadog's APM is the opposite, priced per host, with 150 GB of spans included. A high-traffic service on few hosts can be cheaper on Datadog APM; many small, quiet services are cheaper on Application Signals. Without Transaction Search, Application Signals is billed per signal instead, one per inbound and one per outbound request: (20M + 40M) × $1.50 per million is $90 a month for this startup, with no full span search. Run your own span volume and request count through both before deciding.

Not included: the CPU and memory the collectors use on your hosts, AWS service metrics beyond the free tier, and the engineering time to set it up. This fits next to the bills in how much AWS costs a startup; the CloudWatch Logs line there is the same $0.50 per GB.

For the same startup, Grafana Cloud came to about $375 a month, New Relic to about $348, and a self-hosted open-source stack to about $144 plus engineer time. All three are under a trimmed Datadog at $468. For Sentry, Honeycomb, Dynatrace, SigNoz and the rest, with setup effort for each, see APM and monitoring tools compared for startups.

Grafana Cloud pricing: about $375 a month.

Grafana Cloud Pro has a $19 platform fee and generous included usage: 10,000 active metric series, 50 GB each of logs, traces and profiles. Application Observability has no included host-hours on Pro: the 2,232 host-hours on the pricing page are a Free plan limit, so every APM host-hour is billed. Logs are now billed in parts: $0.05 per GB to process everything you send, $0.40 per GB written above the 50 GB allowance, and a retention charge only beyond the 30 days a paid account includes.

New Relic vs Datadog pricing: a different model, about $348 a month.

New Relic doesn't charge per host. You pay for data ingested above 100 GB a month ($0.40 per GB on the original data option, $0.60 on Data Plus) and for users. Basic users are free, a core user is $49, and on the Standard plan full platform users are $10 for the first and $99 for each additional, up to five. On Pro, a full platform user is $349 a month on an annual commitment.

With New Relic the bill follows people and data volume, so it suits many hosts and few engineers who need the full product. Five full users on Standard would make the users line $406 and the total $546. Check which features basic users can open before you plan around free seats. The 100 GB metrics and events figure is our assumption; New Relic counts every type of data toward ingest, so measure yours in the trial.

Self-hosted, open-source Datadog alternative: about $144 a month, plus your time.

Prometheus, Loki, Tempo and Grafana (or an all-in-one open-source tool) on one m7g.xlarge instance with object storage in S3:

One instance means no high availability: when it's down, so are your alerts, usually at the moment you need them. Upgrades, disk filling up, retention tuning and the monitoring of the monitoring are yours. It also belongs in your disaster recovery plan like any other single instance. Our estimate is four to eight engineer-hours in a quiet month. For a team without an infrastructure person, that time is the real price, and it's usually more than the extra $65 to $204 a month a managed option costs on these numbers.