Datadog's bill has a way of tripling the quarter you actually start using it. Here are eight alternatives we've run in production and what each one costs you.
8 Datadog Alternatives That Cut Your Bill
The Datadog invoice is a rite of passage. It starts small, everyone loves the dashboards, and then someone turns on APM across every service, custom metrics start multiplying, and eleven months later finance forwards you a number with a question mark in the subject line. We've watched a 40-person startup go from $3k a month to $70k a month without adding a single new product line. The usage didn't explode. The instrumentation did.
Datadog is genuinely good software. That's not the problem. The problem is the pricing model punishes exactly the behavior it encourages: more custom metrics, more indexed logs, more hosts, more of everything, each metered separately. Custom metrics alone can quietly become the biggest line item on the bill, and most teams don't find out until they read the breakdown line by line.
So here are eight places teams actually go when they leave, based on migrations we've done or cleaned up after. None of them are free lunches. Each one asks you to accept something in exchange for the smaller bill.
Best at: being the flexible, open-standards hub for metrics, logs, and traces without vendor lock on the data layer.
Pricing model: usage-based on the cloud tier (active series, log GB ingested, trace GB), with a usable free tier. Self-hosted is "free" if you don't count the engineers running it.
The tradeoff: you're assembling a stack, not buying a product. The self-hosted route trades license cost for operational cost, and Mimir and Loki at scale are real systems to run. Grafana Cloud softens that, but you still think in components rather than one polished surface.
Best at: a broad all-in-one platform with a pricing model that confuses fewer people.
Pricing model: you pay for data ingested (GB) plus per-user seats. That's it. No separate meter for custom metrics, hosts, and containers.
The tradeoff: the user-based pricing bites if you want the whole engineering org in the tool, and heavy ingest still adds up. The UI is dense and the query language takes a week to stop fighting.
Best at: large enterprise estates where automatic dependency mapping and root-cause analysis earn their keep.
Pricing model: consumption-based across hosts, ingest, and its analytics engine. Not cheap, and not obviously cheaper than Datadog at first glance.
The tradeoff: you're often not leaving for a lower bill here, you're leaving for the automation. The agent does a lot on its own, which is great until you want to override what it decided on its own.
Best at: an open-source, OpenTelemetry-native all-in-one for teams that want to self-host and own their data.
Pricing model: free if self-hosted; managed cloud is usage-based and undercuts the incumbents meaningfully.
The tradeoff: it's younger, so the ecosystem and integrations are thinner, and self-hosting means you babysit ClickHouse. For teams already committed to OTel, though, it fits like it was built for you, because it was.
Best at: high-cardinality debugging, where you slice by user ID, request ID, or build SHA without pre-aggregating.
Pricing model: event-based, priced on volume of events ingested, with tiers.
The tradeoff: it's not a drop-in dashboards-and-alerts replacement for everything Datadog does. It's a debugging tool with a strong opinion about how you should observe systems, and you re-learn your workflow around events and traces rather than metrics dashboards.
Best at: log-heavy shops that want full-text search and already know the ELK muscle memory.
Pricing model: Elastic Cloud is resource-based (compute plus storage). OpenSearch is the open-source fork you run yourself, so the cost is your cluster.
The tradeoff: running Elasticsearch or OpenSearch well is a skill. Hot-warm-cold tiers, shard sizing, and cluster health become someone's part-time job. Great logs, more operational surface.
Best at: cost control at genuine scale, where the pitch is literally "we shrink your observability bill."
Pricing model: enterprise, based on the volume of data you keep after its control plane drops the metrics nobody queries.
The tradeoff: it's aimed at large cloud-native platforms, not a five-service startup. The whole value is in aggressively pruning cardinality before storage, so you commit to a discipline about what data actually matters. Below a certain scale, the math doesn't work.
Best at: smaller teams that want uptime monitoring, log management, and on-call in one affordable place.
Pricing model: flat, predictable tiers that stay readable as you grow. Others in this bracket include Grafana's OSS stack again and hosted Prometheus setups.
The tradeoff: you give up the deep APM and the enormous integration catalog. For a team that mostly needs "is it up, what did the logs say, who gets paged," that's a fine trade. For deep distributed tracing across 200 services, it isn't.
| Why you're leaving |
Best alternative |
| Custom-metrics bill is out of control |
Chronosphere |
| Want open standards, no data lock-in |
Grafana stack |
| Debugging by high-cardinality fields |
Honeycomb |
| Want to self-host everything on OTel |
SigNoz |
| Log-heavy, search-first workflow |
Elastic / OpenSearch |
| Want one broad platform, simpler pricing |
New Relic |
| Small team, predictable flat bill |
Better Stack |
| Enterprise estate, want more automation |
Dynatrace |
The tool swap is the easy part. The expensive part is re-instrumentation. If your services emit metrics and traces through a Datadog agent and its libraries, ripping that out and wiring in a new SDK across dozens of services is weeks of grinding work, plus a period where you run both in parallel and pay two bills at once. Budget for the overlap.
This is the one place we'll push a specific decision: if you're still on vendor-specific agents, move to OpenTelemetry first, before you pick a destination. Instrument once against the OTel spec and the backend becomes a config change instead of a rewrite. It's the difference between a weekend and a quarter when you switch again, and you will switch again. While you're mapping this out, it's worth reading up on how the different APM tools compare on trace quality, because that's where the real lock-in used to live.
Watch for the parts that don't port cleanly: alert definitions, dashboard JSON, and the monitors people built two years ago and forgot they depend on. Export everything and audit what's actually firing before you assume you need to rebuild it. Half of it is dead.
If you're a small-to-mid team feeling the bill and you have OTel or can get there, SigNoz or the Grafana stack is where we'd start, because you keep your data and your options. If custom metrics are the specific thing eating you alive at scale, Chronosphere exists for exactly that and nothing else. If you're debugging weird production behavior more than you're watching dashboards, Honeycomb changes how you work in a good way. And if you want to leave with the least drama, New Relic's simpler pricing is the softest landing. Whatever you pick, fix the instrumentation before the invoice fixes it for you.
