.png)
Key takeaways
1
AKS bills come from two layers: the cluster tier and the worker node VMs. Standard tier fits production; Free tier fits dev and test.
2
Reservations and Savings Plans both discount node costs, but the right pick depends on whether your VM setup stays fixed or changes. Rightsize before you commit, or you lock in waste at a discount.
3
Spot node pools and cluster start/stop cut costs further, but only for interruption-tolerant workloads and non-continuous schedules. Use them alongside the tier and commitment decisions, not in place of them.
AKS costs have a way of surprising people. You spin up what looks like a small cluster, add a few node pools, leave monitoring enabled, and the month-end bill is four times what you expected.
Compute is the obvious line item, but AKS bills have several moving parts:
Compute is the obvious line item, but AKS bills have several moving parts:
- The cluster management tier
- The VM type and size for each node pool
- The purchase model for those VMs
- Storage and networking
- Monitoring infrastructure
The short answer
Optimizing AKS costs comes down to where the money actually goes: the worker node VMs, not the cluster fee. Run production on the Standard tier for the SLA, keep dev and test clusters on Free, and put your real effort into sizing node pools correctly before buying any commitment.
Committing to Reservations or Savings Plans against an oversized baseline just locks in waste at a discount. Rightsize first, commit only to the stable compute floor, and let Spot node pools and cluster stop/start handle the workloads that don’t need to run at full price around the clock.
AKS cost optimization checklist: run production on Standard and dev/test on Free; set accurate pod resource requests; autoscale through a full usage cycle before committing; commit Reservations or Savings Plans only to the stable compute floor; move interruption-tolerant workloads to Spot node pools; stop non-continuous clusters during off-hours.
Committing to Reservations or Savings Plans against an oversized baseline just locks in waste at a discount. Rightsize first, commit only to the stable compute floor, and let Spot node pools and cluster stop/start handle the workloads that don’t need to run at full price around the clock.
AKS cost optimization checklist: run production on Standard and dev/test on Free; set accurate pod resource requests; autoscale through a full usage cycle before committing; commit Reservations or Savings Plans only to the stable compute floor; move interruption-tolerant workloads to Spot node pools; stop non-continuous clusters during off-hours.
AKS Cost Structure: What You Actually Pay For
AKS has two distinct billing layers that are easy to conflate but important to keep separate.
Cluster management tier charges
| Tier | Control Plane Fee | LTS Support | Best For |
|---|---|---|---|
| Free | $0/cluster | No | Dev/test only |
| Standard | $0.10/hr (~$73/mo) | No | Production workloads |
| Premium | $0.60/hr (~$438/mo) | Yes (24-month LTS) | Regulated or large-scale |
Source: Microsoft Learn AKS pricing tiers and AKS pricing. Verified August 2026.
Why this matters: Running production on the Free tier is a reliability trade-off, not a cost optimization. Long Term Support 24 months of support per Kubernetes version is available only on the Premium tier.
Why this matters: Running production on the Free tier is a reliability trade-off, not a cost optimization. Long Term Support 24 months of support per Kubernetes version is available only on the Premium tier.
Three tier decisions cover most environments:
- Do not run production on the Free tier. It has no financially-backed SLA, and Microsoft recommends it only for clusters with fewer than 10 nodes (it supports up to 1,000).
- Do not run large dev and test clusters on Standard. Non-production clusters do not need uptime guarantees; Free for non-production and Standard for production is the correct split.
- Consider AKS Automatic for new clusters. The fully managed mode runs on the Standard tier and continuously right-sizes infrastructure to actual workload demands. Do not plan on cluster stop/start for AKS Automatic clusters, because the mode uses Node Auto-Provisioning, which that feature does not support.
Worker node VM charges
The second billing layer worker node VMs is far larger than the tier fee. AKS charges for only the nodes attached to the cluster, billed at the same rates as any other Azure VM pay-as-you-go, or discounted through Azure Reservations or Azure Savings Plans. Source: AKS cost management documentation.
Additional charges apply on top of node compute:
Additional charges apply on top of node compute:
- Managed Disk storage for node OS disks
- Public IP addresses for exposed endpoints
- Egress data transfer
- Azure Monitor or Log Analytics costs if container monitoring is enabled
VM Sizing and Node Pool Configuration
VM family, size, and pool structure are the highest-impact optimizations before commitments apply.
Match VM family to workload type
AKS node pools can use most Azure VM types directly:
- General-purpose D-series: balanced workloads
- Compute-optimized F-series: CPU-heavy workloads
- Memory-optimized E-series and M-series: memory-heavy workloads
- GPU-enabled NC/ND/NV series: ML inference and training
Pod resource requests: the sizing lever inside the cluster
Kubernetes schedules pods on resource requests. If a pod requests 4 vCPU and 8 GB but runs at 0.5 vCPU and 1 GB, the cluster provisions capacity for the full request, that headroom translates directly to node count and VM billing. Accurate requests are the most direct cost lever requiring no infrastructure change; the AKS Cost Analysis add-on (below) shows which teams drive request inflation.
Multiple node pools for different workload types
AKS supports multiple node pools with different VM families. Separating distinct profiles, a CPU-optimized pool for processing, a memory-optimized pool for in-memory databases, a Spot pool for batch jobs, lets you price each pool to its workload instead of over-provisioning one pool for the worst case.
Autoscale before you commit
Autoscaling keeps paid capacity matched to actual demand. The Horizontal Pod Autoscaler scales replica counts with load, the Cluster Autoscaler adds and removes nodes as pods need them, and Node Auto-Provisioning selects right-sized VMs for pending pods automatically. Running these for a full usage cycle before purchasing commitments ensures you commit to a genuinely efficient baseline rather than an inflated one. Source: AKS cost optimization best practices.
The Commitment Layer: Savings Plans vs Reservations
The right commitment instrument for AKS nodes depends directly on whether your node configuration is stable or dynamic.
| Signal | Use Azure Reserved VM Instances | Use Azure Savings Plan |
|---|---|---|
| Node VM family changes rarely | Yes | Either works |
| Node VM family changes with K8s upgrades or product releases | No | Yes |
| You run a single VM series per node pool | Yes | Either works |
| You run AKS alongside other Azure compute (VMs, Functions, ACI) | No | Yes, covers all under one commitment |
| You want maximum discount on a known, stable baseline | Yes, up to 72% (3-yr) | No, up to 65% (3-yr) |
| You want flexibility without configuration lock | No | Yes |
| Your AKS node pools use Spot VMs | Neither, Spot is already discounted; commitments do not apply | Neither |
Source: Azure Reserved VM Instances and Azure Savings Plans, verified August 2026.
Why this matters: Teams default to Reservations for the deeper discount, then discover their node configuration changed six months into a three-year term. Reserve only what you are certain will not change; use Savings Plans for the rest.
Why this matters: Teams default to Reservations for the deeper discount, then discover their node configuration changed six months into a three-year term. Reserve only what you are certain will not change; use Savings Plans for the rest.
Reservations and Savings Plans in practice
Azure Reserved VM Instances apply to a specific VM series and region, cutting node costs by up to 72% on a three-year term; one-year terms deliver a smaller discount with less lock-in.
Size flexibility within a series means a D4s_v5 reservation can cover two D2s_v5 instances or half a D8s_v5. For full mechanics, scope, and exchange policy, see our Azure Reservations guide.
Azure Savings Plans for Compute commit to a fixed hourly spend across eligible services VMs, AKS, Azure Databricks, Azure Virtual Desktop for discounts of up to 65%. The discount follows the spend regardless of VM series, size, or region, which suits clusters where node pool VM types change over time.
The sequencing that works:
Size flexibility within a series means a D4s_v5 reservation can cover two D2s_v5 instances or half a D8s_v5. For full mechanics, scope, and exchange policy, see our Azure Reservations guide.
Azure Savings Plans for Compute commit to a fixed hourly spend across eligible services VMs, AKS, Azure Databricks, Azure Virtual Desktop for discounts of up to 65%. The discount follows the spend regardless of VM series, size, or region, which suits clusters where node pool VM types change over time.
The sequencing that works:
Run your AKS clusters for 60–90 days with accurate monitoring enabled.
Identify the stable compute floor, the minimum VM spend that persists across all usage patterns.
Size an Azure Savings Plan or Reservation to that floor.
Leave variable and burst capacity on pay-as-you-go or Spot.
Review commitment coverage quarterly as workload patterns evolve; our Azure commitment management strategy guide covers the review cadence in depth.
Assumptions: 20-node AKS cluster, Standard_D8s_v5 nodes (8 vCPU, 32 GB RAM), East US region, running 730 hours/month, pay-as-you-go baseline.
| Scenario | vs. Pay-as-You-Go |
|---|---|
| Pay-as-you-go (baseline) | — |
| 1-year Azure Reserved VM Instances | ~38% savings |
| 3-year Azure Reserved VM Instances | ~55% savings |
| 30% Spot node pool + 70% 1-yr Reserved | ~55% savings, with eviction risk on 30% |
| Azure Savings Plan (1-year, flexible) | ~33% savings |
Figures are illustrative; confirm current pricing with the Azure Pricing Calculator. Add the Standard tier fee ($73/mo) to each scenario; it does not change across commitment models.
Key takeaway:
A mixed model Reservations for the stable baseline, Savings Plans for the dynamic layer typically delivers the best risk-adjusted outcome.
Spot VMs for AKS: Savings and Eviction Risk
Azure Spot VMs offer unused Azure capacity at up to 90% discount, and you can add Spot node pools for user workloads; the system node pool must stay on-demand.
The catch: Azure can evict Spot nodes with up to 30 seconds’ notice shorter than AWS Spot’s 2-minute warning, so Spot suits only interruption-tolerant workloads like batch jobs, CI/CD builds, and replicated stateless services.
Pair a Spot pool with on-demand pools in the same cluster and let the Cluster Autoscaler fall back when evictions occur. Commitments that do not apply to Spot, it is already discounted. For pricing, eviction policy, and workload fit, see our Azure Spot VMs guide.
The catch: Azure can evict Spot nodes with up to 30 seconds’ notice shorter than AWS Spot’s 2-minute warning, so Spot suits only interruption-tolerant workloads like batch jobs, CI/CD builds, and replicated stateless services.
Pair a Spot pool with on-demand pools in the same cluster and let the Cluster Autoscaler fall back when evictions occur. Commitments that do not apply to Spot, it is already discounted. For pricing, eviction policy, and workload fit, see our Azure Spot VMs guide.
Off-Hours Scheduling with AKS Cluster Start/Stop
AKS provides a cluster start and stop feature that shuts down all system and user node pools so you don’t pay for extra compute, while cluster state is preserved for the restart. It fits clusters that need to exist but not run continuously:
- Dev environments used only during business hours
- Staging clusters used for periodic testing
- QA environments used for scheduled test runs
Caution:
stop/start is best suited to noncritical, Virtual Machine Scale Set-backed clusters. It is not supported on clusters using Node Auto-Provisioning (including AKS Automatic clusters), the stop process deletes standalone pods not managed by a controller, and restart in a capacity-constrained region is not guaranteed test restart and workload recovery before relying on it.
A cluster running 8 hours per business day uses roughly a quarter of a calendar month’s hours, cutting its node VM bill by roughly three-quarters.
Stops and starts can be automated via Azure DevOps, Azure Automation, or Logic Apps. Persistent resources like managed disks and public IPs still bill while stopped, but the dominant VM cost does not.
Stops and starts can be automated via Azure DevOps, Azure Automation, or Logic Apps. Persistent resources like managed disks and public IPs still bill while stopped, but the dominant VM cost does not.
Azure Hybrid Benefit for AKS Windows Node Pools
Windows node pools cost more than Linux ones because the node VM rate bundles Windows Server licensing. Azure Hybrid Benefit waives that licensing component leaving only base compute, if your organization holds qualifying Windows Server licenses with active Software Assurance or a qualifying subscription.
Two things make this worth checking. First, if you pay Software Assurance anyway for compliance or support reasons, the incremental cost of applying the benefit is zero.
Second, it stacks with commitments: Hybrid Benefit can be combined with Azure Reserved VM Instances on the same nodes, so the licensing saving and the compute discount apply together. Linux node pools are unaffected. Source: Azure Hybrid Benefit for AKS and Hybrid Benefit pricing.
Two things make this worth checking. First, if you pay Software Assurance anyway for compliance or support reasons, the incremental cost of applying the benefit is zero.
Second, it stacks with commitments: Hybrid Benefit can be combined with Azure Reserved VM Instances on the same nodes, so the licensing saving and the compute discount apply together. Linux node pools are unaffected. Source: Azure Hybrid Benefit for AKS and Hybrid Benefit pricing.
AKS Cost Analysis: Namespace and Workload Attribution
The AKS Cost Analysis add-on provides cost allocation at the namespace and workload level, built on OpenCost and integrated into Azure Cost Management, so you can answer ‘which team’s namespace is driving 60% of our cluster cost’ directly from the Azure portal.
Two constraints: it requires a Standard or Premium tier cluster, and it supports approximately 7,000 containers per cluster based on the current memory limit. The add-on itself has no separate charge. Source: AKS cost analysis documentation.
Two constraints: it requires a Standard or Premium tier cluster, and it supports approximately 7,000 containers per cluster based on the current memory limit. The add-on itself has no separate charge. Source: AKS cost analysis documentation.
How Usage.ai Fits Into AKS Cost Management
Usage.ai handles the commitment layer for AKS, the Azure Savings Plan and Reservation purchasing that determines what you pay per node VM hour. It is the sequencing above, run for you rather than managed by hand. Under the Flex Commitment Program, the platform:
Every Flex Commitment carries cashback protection. If a commitment ends up costing more than the on-demand equivalent for the same usage, that difference returns as cashback in real money. The fee is a percentage of realized savings only: no savings, no fee.
- Analyzes your cloud usage data and recommends the commitments worth purchasing
- Purchases approved commitments through the Azure API nothing is bought without your approval
- Handles monitoring and optimization from there, with a monthly savings report across every connected account
Every Flex Commitment carries cashback protection. If a commitment ends up costing more than the on-demand equivalent for the same usage, that difference returns as cashback in real money. The fee is a percentage of realized savings only: no savings, no fee.
Frequently asked questions
What does AKS actually charge for?
A cluster management tier fee (none on Free; hourly on Standard and Premium) plus worker node VM charges at standard Azure VM rates. Storage, public IPs, egress, and monitoring add smaller line items.
Should I use AKS Free tier or Standard tier?
Standard for all production: it carries a financially-backed SLA (99.95% with availability zones, 99.9% without) and enables Cost Analysis. Free is for dev and test only, with no financially-backed SLA.
What is the difference between Azure Savings Plans and Reservations for AKS nodes?
Reservations lock to a VM series and region for up to 72% savings, with size flexibility within the series. Savings Plans commit to hourly spend across eligible services for up to 65%, following the spend when VM types change.
How do Azure Spot VMs work in AKS?
Spot node pools access unused Azure capacity at up to 90% discount, but Azure can reclaim Spot VMs with up to 30 seconds' notice. Use them for batch jobs, CI/CD builds, and replicated stateless services not stateful production workloads.
What is AKS Automatic?
A fully managed cluster mode, generally available since 2025, that automates node provisioning, scaling, upgrades, and networking on the Standard tier, continuously right-sizing infrastructure to actual pod requirements.
Share
Facebook
X
LinkedIn
Reddit
