Per-host vs per-GB monitoring pricing: which model punishes your architecture

Quick answer

Per-host pricing (roughly $15 to $31 per host-month) charges for machines regardless of telemetry volume; per-GB pricing (roughly $0.10 to $2.76 per GB depending on platform and tier) charges for data regardless of machine count. Divide your monthly telemetry GB by your host count to find the break-even: at a $31 host price and $0.50 per GB, the crossover is 62 GB per host per month, about 2 GB per host per day. Below that, per-host is more expensive; above it, per-host is a bargain. Container density and serverless host-counting rules move the line sharply.

Two pricing models dominate observability, and they are close to opposites. One charges per machine and ignores how much data each sends. The other charges per byte and ignores how many machines sent it. Your infrastructure shape determines which one treats you well, and that shape is something you control, which makes this a design question rather than just a procurement one.

Finding your break-even

The calculation is one division. Take monthly telemetry volume in GB, divide by host count, and compare the result against the host price divided by the per-GB price.

Host price	Per-GB price	Break-even GB per host-month
$31	$0.50 (CloudWatch Logs)	62 GB
$31	$0.25 (Infrequent Access)	124 GB
$31	$2.76 (Azure analytics)	11.2 GB
$15	$0.50	30 GB
$15	$0.10 (cheap SaaS ingest)	150 GB

Sixty-two GB per host-month is about 2 GB per host per day. A busy application server easily exceeds that. A batch worker or a database replica almost never does. So the answer is usually different for different parts of the same estate, which is the first useful insight: the question is not which model to buy, it is which fleets belong under which model.

Where per-host pricing gets brutal

High host count with low telemetry is the punishing case. A 900-instance Spot fleet running batch jobs that each log 200 MB per day produces 180 GB/day of logs, 5,400 GB per month. Per-GB at $0.50 that is $2,700. Per-host at $31 it is $27,900, more than ten times higher, for the same monitoring.

Container density makes it worse in a specific way. If a vendor counts each container as a host or a fraction of one, a node running 60 pods can bill as considerably more than one host. Read the definition carefully: some contracts count containers up to a limit per host and charge per container above it, often around $1 per additional container per month. A 250-node cluster at 60 pods per node with a free allowance of 10 containers per host means 12,500 chargeable containers, $12,500 per month on top of the $7,750 for the nodes.

Autoscaling and short-lived instances add another wrinkle. Per-host billing is usually computed on a high water mark or a percentile of concurrent hosts over the month, so a fleet that scales from 100 to 800 during a daily peak may bill closer to 800 than to its 250-host average. Check whether the contract uses the 99th percentile, the maximum, or a true average, because the difference on a spiky fleet is often 3x.

Where per-GB pricing gets brutal

Few hosts with heavy telemetry is the mirror case. A 25-node cluster running chatty services that produce 400 GB/day of logs generates 12,000 GB per month. Per-host at $31 that is $775. At CloudWatch's $0.50 per GB it is $6,000, and on Azure Monitor analytics logs at roughly $2.76 per GB it is $33,120. Per-GB pricing turns log verbosity directly into invoice line items, which is either a useful incentive or a tax on debugging depending on your view.

Per-GB also punishes high-cardinality metrics indirectly, since more series means more bytes, and it punishes debug logging in production savagely. Under per-host pricing, a team that leaves DEBUG on pays nothing extra; under per-GB, the same mistake can add thousands of dollars per month. See volume reduction for the countermeasures.

The hybrid that most large estates land on

Fleet	Shape	Best model
Customer-facing API nodes	Few hosts, heavy telemetry	Per-host
Batch and data processing	Many hosts, light telemetry	Per-GB
Kubernetes control plane	Few hosts, moderate	Per-host
Edge and CDN logs	No hosts, huge volume	Per-GB with cheap tier
Serverless functions	Host count undefined	Per-GB, usually

Splitting by fleet is more work operationally, because you run two agents and two sets of dashboards, and that operational tax is real. But the cost delta on a large estate frequently exceeds $20,000 per month, which buys a lot of tolerance for two dashboards.

Questions to settle before signing

How is a host defined for Fargate tasks, Lambda functions, and Kubernetes pods? Is the monthly host count an average, a percentile, or a maximum? Are containers counted separately above an allowance, and at what price? Does the host price include APM, profiling, and log management, or are those separate SKUs on the same host count? Is there a commitment discount, and what happens if you undershoot the commitment? What does an overage cost relative to the committed rate?

The last one matters most, because monitoring contracts are usually sized on today's fleet and fleets grow. An overage rate at 1.5x the committed price on a fleet that grows 40% in a year turns a good deal into a bad one. Model the growth. And since the fleet itself is Terraform, price it from the plan against the resource catalog so the host count driving your monitoring contract is a forecast you can run rather than a number you discover.

FAQ

How do I choose between per-host and per-GB monitoring pricing?

Divide your monthly telemetry volume in GB by your host count, then compare against the host price divided by the per-GB price. At a $31 host price and $0.50 per GB, the break-even is 62 GB per host-month, about 2 GB per host per day. Below that, per-host pricing costs more; above it, per-host is cheaper. The answer usually differs across fleets within the same organization.

When is per-host monitoring pricing expensive?

When you have many hosts producing little telemetry. A 900-instance batch fleet logging 200 MB per host per day produces 5,400 GB per month, costing $2,700 at $0.50 per GB but $27,900 at $31 per host. Container counting makes it worse: a 250-node cluster at 60 pods per node with a 10-container allowance can add 12,500 chargeable containers at roughly $1 each on top of the node charge.

When is per-GB monitoring pricing expensive?

When few hosts produce heavy telemetry. A 25-node cluster generating 400 GB/day produces 12,000 GB per month, costing $775 at $31 per host but $6,000 at CloudWatch's $0.50 per GB and $33,120 on Azure Monitor analytics logs at roughly $2.76 per GB. Per-GB pricing also turns mistakes like leaving DEBUG logging on in production directly into thousands of dollars of additional monthly spend.

How does autoscaling affect per-host monitoring bills?

Significantly, depending on the contract. Per-host billing is often computed from a high water mark or a high percentile of concurrent hosts rather than a true average. A fleet scaling from 100 to 800 instances during a daily peak with a 250-host average may bill closer to 800 under a maximum or 99th-percentile clause, roughly 3x the average-based cost. Confirm the measurement method before signing.

Should I use both pricing models across my estate?

Large estates often do. Customer-facing nodes with heavy telemetry and few hosts suit per-host pricing; batch and data processing fleets with many hosts and light telemetry suit per-GB; edge and CDN logs with no hosts at all belong on a cheap per-GB tier. The operational cost is running two agents and two dashboard sets, which is usually worth it when the delta exceeds $20,000 per month.

How does C3X help with monitoring contract decisions?

Your host count is a property of your Terraform: autoscaling group sizes, node group configurations, and task definitions. C3X prices that infrastructure from the plan, which lets you forecast the host count that will drive a per-host contract as the fleet grows, rather than sizing a multi-year agreement on today's number and discovering overage rates when the fleet expands 40%.

What to do next

Forecast the fleet that drives your monitoring contract. C3X reads your Terraform and prices instances and node groups against a live catalog. Start with the quickstart.