What if the sticker price on your monitoring tool is only half the story? That is the uncomfortable truth anyone comparing infrastructure monitoring prices runs into within about ten minutes of opening a vendor’s pricing page. The headline number looks reasonable. Then you read the footnotes about custom metrics, log ingestion, retention windows, and API calls, and the budget you sketched out quietly doubles.
This guide breaks down how the major pricing models actually work, where the hidden costs hide, and how to compare tools on a basis that reflects what you will really spend. No sales pitches, no invented figures. Where prices are illustrative, they are labelled as such, because list prices shift constantly and every number here needs checking against current vendor documentation before it lands in a budget.
Why Headline Prices Mislead
Most infrastructure monitoring vendors do not price the same thing. Some charge per host per month. Others charge per custom metric. Some charge per seat, others per gigabyte ingested or per million API requests. Comparing a per-host tool against a per-metric tool on “starting price” is like comparing rent per apartment against rent per square metre and wondering why the numbers do not line up.
The practical consequence: a tool that looks expensive at the entry tier can be cheaper at scale, and vice versa. A ten-host startup and a two-thousand-host enterprise should not be running the same comparison spreadsheet, because the cost curves cross at different points.
There is a second layer too. Many modern platforms bundle infrastructure monitoring with APM, logs, and tracing. Comparing a standalone infrastructure tool against a bundled observability suite on infrastructure price alone understates the suite’s total value, and comparing them on total price understates the standalone tool’s focus. The scope of the comparison has to be stated explicitly before any number means anything.
The Four Pricing Models, Explained

Per-Host or Per-Agent Pricing
This is the classic model. You pay a monthly fee for each server, virtual machine, or agent reporting in. It is predictable, easy to forecast, and easy to explain to finance.
The catch is that it scales linearly with fleet size. For a 1,000-host estate, a published infrastructure tier in the roughly $15 to $23 per host per month range implies something in the order of $180,000 to $276,000 per year before you add logs, APM, or custom metrics. Treat those figures as illustrative only, since list prices change and enterprise discounts are rarely published, but the arithmetic shows why fleet size dominates the conversation.
Per-host pricing also punishes certain architectures. Ephemeral containers, autoscaling groups, and short-lived batch jobs can churn through “hosts” in ways that make the bill unpredictable unless the vendor handles them specially.
Per-Metric and Per-Custom-Metric Pricing
Some platforms charge by the number of metrics you store, with a separate and often steeper rate for custom metrics. This model aligns cost with usage, which sounds fair until you discover what high-cardinality data does to your bill.
A single metric tagged with user ID, request path, region, and status code can explode into millions of unique time series. Cardinality control is a documented cost-management practice for exactly this reason. The practical lesson: if you choose a per-metric vendor, budget time for tagging discipline before you budget money for the invoice.
Per-Data-Volume Pricing
Log and trace backends commonly charge by ingest volume (per gigabyte) and sometimes by retention. This model is transparent in principle and terrifying in practice, because log volume grows quietly. A debug logging level left on over a long weekend can cost more than a month of infrastructure monitoring.
The standard mitigations are sampling, filtering at the agent, and tiered retention. All of them work. None of them happen automatically unless someone owns the problem.
Per-Seat Pricing
Less common for pure infrastructure monitoring, but frequent in adjacent tooling such as dashboards and on-call management. Per-seat pricing is friendly to large server fleets and hostile to large engineering organisations. If your comparison mixes per-host infrastructure tools with per-seat alerting tools, separate the two lines in your spreadsheet.
Open Source: Free Licence, Real Costs
Prometheus and the Grafana stack (Grafana, Loki, Tempo, Mimir) are open source. Grafana moved its core to AGPLv3 in April 2021. Prometheus is a graduated project under the CNCF umbrella. The licence fee is genuinely zero, and that is a real advantage, not a marketing trick.
But “free” is where the accounting gets interesting. Prometheus’s own documentation is refreshingly blunt about its limits: it is not intended as long-term storage, and its local storage is not clustered or replicated. Long-term retention requires remote-write to a compatible backend. High availability requires running duplicate scrapers or a dedicated HA layer.
So the self-hosted cost breaks into four buckets:
| Cost bucket | What it covers | Why it surprises people |
|---|---|---|
| Compute | Prometheus servers, remote-write backends, Grafana instances | Grows with retention and cardinality, not host count |
| Storage | Metrics TSDB, log store, object storage for long-term retention | Retention windows silently multiply disk needs |
| Engineering time | Upgrades, scaling, backup, capacity planning | Rarely counted, usually the largest line |
| On-call | Someone to fix it at 3 a.m. | The tool monitors everything except itself |
The last two are the ones teams forget. A self-hosted stack is not a product you buy; it is a system you operate. That is fine if you have platform engineers with spare capacity and a genuine reason to own the stack. It is a poor trade if you are a five-person team whose actual job is something else.
Managed open-source offerings sit in the middle: same open formats, someone else runs the storage and HA, and you pay for that operational relief rather than for a licence. That is often the most honest comparison point against a proprietary per-host vendor.
Cloud-Native Monitoring: Bundled, Not Free at Scale
AWS CloudWatch, Azure Monitor, and Google Cloud Monitoring feel free because they come with the platform. They are not free at scale. They bill per metric, per alarm, per API request, and per gigabyte of logs ingested, with free tiers that cover small workloads and stop covering them quickly.
The relevant comparison point is therefore not the list price but the free-tier boundary. Where does your workload cross it, and what does the bill look like on the other side? For a small deployment, native monitoring is often the cheapest sensible option because the integration cost is near zero. For a large, multi-cloud estate, the per-metric and per-request charges accumulate in ways that make a dedicated tool look reasonable again.
One practical warning: native monitoring is sticky. Dashboards, alarms, and IAM policies accumulate around it. Migrating away later costs more than choosing it cost in the first place, so make the choice with a two-year horizon in mind, not a two-week one.
The Costs Nobody Puts on the Pricing Page
Here is where real comparisons are won and lost. Beyond licence or subscription fees, every option carries costs that rarely appear in a pricing table.
Cardinality management. High-cardinality metrics inflate per-metric bills and slow self-hosted storage. Tagging standards, metric naming conventions, and periodic cardinality audits are ongoing work, not one-off tasks.
Retention trade-offs. Long retention is where costs concentrate. Most teams need days of high-resolution data and months of downsampled data. Vendors that charge for retention reward this pattern; vendors that do not still pay for it in storage.
Integration and instrumentation. Getting data in is work. OpenTelemetry has become the common standard for metrics, traces, and logs, and adopting it reduces lock-in, but it does not eliminate the engineering effort of instrumenting services properly.
Alert quality. A tool that generates noise costs engineering hours in a way no invoice captures. Alert fatigue is a real operational expense, and it is usually worse in tools that were chosen on price alone.
Migration cost. Moving monitoring platforms means rebuilding dashboards, re-tuning alerts, and re-training everyone who reads them. It is the single most underestimated line in any comparison, and the reason “we will switch later if it gets expensive” rarely happens.
How to Build a Comparison That Holds Up
The method matters more than the spreadsheet template. A comparison that survives contact with reality follows a few rules.
- Fix the scope first. Decide whether you are comparing infrastructure monitoring alone or a full observability suite. Write it down. Every subsequent number depends on it.
- Model your actual fleet, not a round number. Use real host counts, real container churn, real log volume, and real metric cardinality from the last ninety days.
- Price three scenarios. Current state, twelve months out, and a stress case at double the fleet. The stress case is where pricing models diverge most.
- Add operational cost explicitly. Estimate engineering hours for self-hosted options at a real loaded rate. If nobody can estimate it, that is itself a finding.
- Check free-tier boundaries. For cloud-native tools, find the exact point where free becomes paid and see how close you sit to it.
- Re-verify every figure. List prices change, discounts are unpublished, and currency and region shift the numbers. Any figure older than a few months is a rumour.
- Weight alert quality and workflow fit. A cheaper tool that your team ignores is the most expensive option on the list.
A useful discipline: write the one-paragraph justification for your chosen option before you build the spreadsheet. If the paragraph is about price alone, the comparison is probably incomplete.
Where Each Model Fits
| Situation | Likely best fit | Why |
|---|---|---|
| Small team, single cloud, modest fleet | Cloud-native monitoring | Integration cost near zero; free tier covers the workload |
| Large stable fleet, predictable growth | Per-host pricing | Costs forecast cleanly; discounts negotiable at volume |
| Highly dynamic container estate | Per-metric with strong cardinality control, or managed open source | Per-host models struggle with churn |
| Log-heavy workloads | Volume-based with sampling and tiered retention | Ingest discipline matters more than vendor choice |
| Team with platform engineers and a real reason to own the stack | Self-hosted Prometheus and Grafana stack | Licence cost zero; operational cost is the trade |
| Regulated or multi-cloud estate | Managed open source or bundled suite | Portability and auditability outweigh unit price |
None of these are universal answers. They are starting positions for a comparison, which is the most any pricing guide can honestly offer.
Conclusion
The pricing comparison question has a slightly annoying answer: the cheapest infrastructure monitoring tool depends almost entirely on your fleet shape, your growth curve, and how much operational work your team can genuinely absorb. Per-host pricing rewards stable, predictable estates. Per-metric pricing rewards disciplined tagging. Volume pricing rewards log hygiene. Open source rewards teams with platform engineering capacity and punishes everyone else with hidden operational debt.
What actually works in practice is boring and effective: fix the scope, model real numbers, price a stress scenario, count engineering hours honestly, and re-verify every figure before signing. What does not work is comparing headline prices across models that measure different things, or assuming “free” and “cheap” mean the same thing. They rarely do.
FAQ
What is the cheapest infrastructure monitoring option overall?
There is no universal cheapest option, because pricing models measure different things. For a small deployment on a single cloud, native monitoring is often cheapest because the free tier covers the workload and integration cost is near zero. For a large stable fleet, per-host pricing with a negotiated discount frequently wins. For teams with platform engineers, self-hosted open source has zero licence cost but real operational cost. The cheapest option is the one whose cost model matches your fleet shape.
Why do monitoring bills exceed the quoted price?
Three reasons dominate. First, custom metrics and high-cardinality tags inflate per-metric charges far beyond the base tier. Second, log ingestion volume grows quietly and is billed per gigabyte. Third, retention windows multiply storage requirements. Vendors quote entry-tier prices for simple workloads; most real environments are not simple. Budget for cardinality control and log filtering from day one, not after the first surprising invoice.
Is Prometheus free for long-term monitoring?
Prometheus is open source with no licence fee, but its own documentation states it is not intended as long-term storage, and its local storage is not clustered or replicated. Long-term retention requires remote-write to a compatible backend, and high availability requires additional architecture. So the software is free, but the storage, compute, and engineering time to run it at scale are not. The honest framing is “free licence, real operating cost.”
How does per-host pricing scale for a large fleet?
Linearly, which is the point and the problem. A published infrastructure tier in the roughly $15 to $23 per host per month range, applied to 1,000 hosts, implies something like $180,000 to $276,000 per year before logs, APM, or custom metrics. Those figures are illustrative and must be verified, and enterprise discounts are rarely published. The takeaway is structural: fleet size, not feature list, dominates cost under per-host models.
Do cloud provider monitoring tools cost extra?
Yes, beyond their free tiers. Cloud-native monitoring services bill per metric, per alarm, per API request, and per gigabyte of logs ingested. Small workloads often stay inside the free tier, which makes the tools feel free. Larger workloads cross that boundary quickly. The practical step is to find the exact free-tier limit for your usage pattern and model the bill on the other side before committing.
How should a team compare monitoring tools fairly?
Fix the scope first (infrastructure only versus full suite), then model real fleet numbers from the last ninety days rather than round figures. Price three scenarios: current, twelve months out, and double the fleet. Add engineering hours for self-hosted options at a real loaded rate. Check free-tier boundaries. Re-verify every price against current documentation. Finally, weight alert quality and workflow fit, because a cheap tool the team ignores costs more than an expensive one they use.