Kubernetes cost intelligence

See inside every cluster, down to the pod.

Q: How does CarnotCloud collect Kubernetes data?

A lightweight, read-only collector installs with one Helm chart and sends resource requests, usage and labels — never application data or secrets. It can also read from your existing Prometheus.

Q: Does CarnotCloud change our clusters automatically?

CarnotCloud is read-only by design. It gives engineers precise, evidence-backed recommendations and tracks the savings once changes ship through your normal deployment process — so your team stays in full control of production.

Q: Which distributions are supported?

Amazon EKS, Azure AKS, Google GKE, Red Hat OpenShift, Rancher and any CNCF-conformant self-managed cluster, plus ECS and Fargate.

Q: How are idle and shared costs handled?

You choose: distribute idle and system overhead proportionally to requests, to usage, evenly, or keep them in a dedicated platform bucket. Every rule is transparent and auditable.

Kubernetes is where cloud cost goes to hide. CarnotCloud reconciles every node to your real bill, allocates it to namespaces, workloads and teams, exposes idle capacity — and tells engineers exactly which requests to change.

Book a demo Free savings assessment

Read-only access Live in 15 minutes SOC 2 Type II

Cluster: prod-eu-west-1 (EKS)Monthly cost by namespace · incl. shared & idle

$126K/mo

checkout $41.2K
search $28.7K
ml-inference $22.3K
platform $14.1K
Idle & unallocated $19.8K

WorkloadCPU requestedCPU p95 usedSavings/mo

checkout-api8.0 vCPU2.1 vCPU$6,420

search-indexer16.0 vCPU6.8 vCPU$4,980

recs-worker4.0 vCPU0.9 vCPU$2,310

<5%

unallocated cluster cost after shared and idle distribution

30–45%

typical over-provisioning uncovered in CPU and memory requests

1 Helm chart

lightweight, read-only collector — live in minutes

Capabilities

What you get with Kubernetes Cost Intelligence

Cluster → pod allocation

Cost by cluster, namespace, deployment, pod, container and label — on EKS, AKS, GKE, OpenShift, Rancher and self-managed clusters.

Bill-accurate node costs

Node costs reconciled to your invoice, including Savings Plans, RIs, Spot and enterprise discounts — not list prices.

Idle & shared cost distribution

Distribute idle capacity, kube-system, monitoring and control-plane overhead with transparent, configurable rules.

Request rightsizing

p95/p99-based CPU and memory request and limit recommendations per workload, with risk scoring and copy-ready YAML.

Node & instance guidance

Node-group and instance-family recommendations (including Graviton and Spot suitability) based on real bin-packing efficiency.

GPU visibility

Per-pod GPU allocation and utilization for training and inference clusters, so idle accelerators never hide.

Allocation teams trust

End the “whose cluster is this?” debate

Shared clusters make ownership fuzzy. CarnotCloud maps every workload to a team, product or customer and shows each one its fair share of the cluster — including the idle capacity it reserved but didn’t use.

Namespace and label mapping to your business hierarchy
Efficiency score per workload and per team
Showback-ready reports for platform teams
Out-of-cluster costs (storage, load balancers, data transfer) attached to workloads

Cluster: prod-eu-west-1 (EKS)Monthly cost by namespace · incl. shared & idle

$126K/mo

checkout $41.2K
search $28.7K
ml-inference $22.3K
platform $14.1K
Idle & unallocated $19.8K

WorkloadCPU requestedCPU p95 usedSavings/mo

checkout-api8.0 vCPU2.1 vCPU$6,420

search-indexer16.0 vCPU6.8 vCPU$4,980

recs-worker4.0 vCPU0.9 vCPU$2,310

Recommendations engineers accept

Rightsizing advice with evidence attached

Every recommendation shows the utilization history, headroom and risk behind it. Engineers review, apply in their own pipeline and watch realized savings appear — with full control over every change.

Workload-level recommendations with p95/p99 headroom
Copy-ready YAML and Helm values diffs
One-click Jira and GitHub issues with context
Realized-savings tracking after changes ship

Optimization opportunitiesRanked by savings ÷ effort

$94K/mo

RecommendationMonthly savingsEffort

m6i.4xlarge → m7g.2xlarge142 instances · EC2$38,240Low Jira

Unattached EBS volumes1,204 volumes · us-east-1$12,910Low Jira

Idle RDS: analytics-stagingdb.r6g.8xlarge · 0.4% CPU$9,870Low Jira

S3 → Intelligent-Tiering3.1 PB · 41 buckets$18,700Medium Jira

Azure VM scale set rightsizingD16s_v5 → D8s_v5 · 64 VMs$14,350Medium Jira

38% lower Kubernetes spend

“Pod-level allocation ended the “whose cluster is this?” debate. Following the request-rightsizing recommendations took 38% off our EKS bill without a single incident.”

Sofia T.Platform Engineering Lead, Streaming media company

FAQ

Kubernetes Cost Intelligence: common questions

How does CarnotCloud collect Kubernetes data?

A lightweight, read-only collector installs with one Helm chart and sends resource requests, usage and labels — never application data or secrets. It can also read from your existing Prometheus.

Does CarnotCloud change our clusters automatically?

CarnotCloud is read-only by design. It gives engineers precise, evidence-backed recommendations and tracks the savings once changes ship through your normal deployment process — so your team stays in full control of production.

Which distributions are supported?

Amazon EKS, Azure AKS, Google GKE, Red Hat OpenShift, Rancher and any CNCF-conformant self-managed cluster, plus ECS and Fargate.

How are idle and shared costs handled?

You choose: distribute idle and system overhead proportionally to requests, to usage, evenly, or keep them in a dedicated platform bucket. Every rule is transparent and auditable.

Keep exploring

See Kubernetes Cost Intelligence on your own data.

Book a 30-minute demo tailored to your stack, or start with a free savings assessment — read-only, no agents, no obligation.

Book a demo Get a free savings assessment

Read-only, agentless setup
First insights in 15 minutes
SOC 2 Type II & ISO 27001