Skip to main content
Cloud DevOps Solutions
Get a quote Talk to an engineer
Cloud Cost 2026-11-12 14 min read

Azure Cost Optimisation: 10 Proven Ways to Cut Your Azure Bill

Saddique Khan

DevOps / SRE Engineer & CEO, Cloud DevOps Solutions

Azure bills have a habit of growing faster than the workloads that generate them. A subscription that started at $5,000/month becomes $25,000 once you add production redundancy, managed databases, and a few “temporary” resources that nobody cleaned up.

The good news: Azure provides more built-in cost optimisation tools than most teams realise. The challenge is knowing which levers to pull first. This guide covers 10 techniques we apply on Azure engagements, roughly ordered by impact.

1. Right-size your virtual machines

VM compute is the largest line item on most Azure bills. The default behaviour is to provision larger than necessary because nobody wants to be the person who chose too small and caused a production issue. The result: VMs running at 5-15% average CPU utilisation.

How to find oversized VMs

Azure Advisor flags VMs with consistently low utilisation. Navigate to Advisor > Cost > Right-size or shut down underutilised virtual machines. Advisor analyses 14 days of CPU, memory (via Azure Monitor Agent), network, and disk metrics.

Azure Monitor gives you the raw data. Check these metrics for each VM:

  • Average CPU over 14 days (below 20% is a candidate for downsizing)
  • P95 CPU (the peak that matters for user experience)
  • Memory utilisation (requires Azure Monitor Agent or Diagnostics Extension)
  • Network throughput relative to the VM’s network bandwidth cap

How to right-size

  1. Identify VMs where P95 CPU is below 40% and average is below 15%
  2. Find the next smaller VM size in the same family (e.g. Standard_D4s_v5 to Standard_D2s_v5)
  3. Verify the smaller size meets memory, disk, and network requirements
  4. Resize during a maintenance window (Azure VMs restart during resize)

A D4s_v5 costs roughly $140/month. A D2s_v5 costs $70. If you have 20 oversized VMs, that single change saves $1,400/month.

Automated right-sizing

For AKS clusters, consider the Vertical Pod Autoscaler (VPA) for container right-sizing and cluster autoscaler for node pool scaling. For standalone VMs, Azure Automanage can apply Advisor recommendations automatically.

2. Use Reserved Instances and Savings Plans

On-Demand pricing is the most expensive way to run Azure workloads. If a VM will run for 12 months or more, a commitment discount saves 30-72% depending on the term and payment option.

Reserved Instances (RIs)

RIs lock in a specific VM size and region for one or three years:

TermPaymentTypical saving vs On-Demand
1 year, monthlyMonthly30-40%
1 year, upfrontOne payment35-42%
3 year, monthlyMonthly55-62%
3 year, upfrontOne payment60-72%

RIs apply automatically to matching VMs. Instance size flexibility within the same series means a D4s_v5 RI covers two D2s_v5 instances or one D4s_v5.

Azure Savings Plans

Savings Plans (launched in 2022) provide commitment-based discounts that are more flexible than RIs. You commit to a fixed hourly spend (e.g. $10/hour) for one or three years, and the discount applies to any eligible compute across regions and VM sizes.

When to use RIs vs Savings Plans:

  • RIs: workloads locked to a specific region and VM family (databases, control planes)
  • Savings Plans: dynamic workloads that may shift regions or VM sizes (AKS node pools, batch processing)

How to calculate

Azure Cost Management includes a Reservations blade with recommendations based on your actual usage. Start there. For a more detailed analysis, export your usage data and model scenarios: what if you reserve 70% of your baseline and leave 30% On-Demand for burst?

3. Shut down dev/test environments outside business hours

Development, staging, and QA environments typically run 24/7 but are used only during business hours: roughly 10 hours per day, 5 days per week. That is 50 hours of use and 118 hours of waste.

Azure Automation runbooks

Create an Azure Automation account with two runbooks:

  • Start-VMs: runs at 08:00 on weekdays, starts tagged VMs
  • Stop-VMs: runs at 18:00 on weekdays, deallocates tagged VMs

Tag your dev/test VMs with auto-shutdown: true and reference that tag in the runbook logic.

Saving: shutting down VMs for 70% of the week saves 70% of their compute cost. A $5,000/month dev environment becomes $1,500.

Azure DevTest Labs

For more structured environments, Azure DevTest Labs provides built-in auto-shutdown policies, formula-based VM creation, and cost tracking per lab. It also offers DevTest pricing (up to 50% off) for eligible MSDN subscribers.

AKS node pools

For AKS clusters used only during business hours, scale the user node pool to zero nodes outside hours using the cluster autoscaler’s --min-count 0 setting, or use the AKS Start/Stop Cluster feature for entire non-production clusters.

4. Optimise storage tiers

Azure Storage offers four access tiers. Putting data in the wrong tier is one of the most common sources of waste:

TierStorage cost (per GB/month)Access costBest for
Hot~$0.018LowFrequently accessed data
Cool~$0.010MediumData accessed less than once a month
Cold~$0.0036HigherData accessed less than once a quarter
Archive~$0.002High + rehydration delayData accessed less than once a year

Lifecycle management policies

Create lifecycle management rules to automatically move blobs between tiers based on last access time:

{
  "rules": [
    {
      "name": "move-to-cool-after-30-days",
      "type": "Lifecycle",
      "definition": {
        "actions": {
          "baseBlob": {
            "tierToCool": {
              "daysAfterLastAccessTimeGreaterThan": 30
            },
            "tierToArchive": {
              "daysAfterLastAccessTimeGreaterThan": 180
            }
          }
        },
        "filters": {
          "blobTypes": ["blockBlob"],
          "prefixMatch": ["logs/", "backups/"]
        }
      }
    }
  ]
}

Enable last access time tracking on the storage account for these rules to work.

Managed disk tiers

Review your managed disks. Premium SSD (P-series) disks are expensive but necessary for production databases. Dev/test workloads often run fine on Standard SSD (E-series) or even Standard HDD (S-series). A P30 (1 TiB Premium SSD) costs $122/month; an E30 (1 TiB Standard SSD) costs $61.

Also check for unattached managed disks. When a VM is deleted, its disks are not deleted by default. Run:

az disk list --query "[?managedBy==null].[name, diskSizeGb, sku.name]" -o table

5. Apply Azure Hybrid Benefit

If your organisation has Windows Server or SQL Server licences with Software Assurance (or qualifying subscription licences), Azure Hybrid Benefit lets you use those licences on Azure VMs. The saving is substantial:

  • Windows Server: up to 40% off VM cost (covers the Windows licence portion)
  • SQL Server: up to 55% off Azure SQL or SQL Server on VMs
  • Combined: Windows Server + SQL Server Hybrid Benefit stacks, saving up to 80% compared to pay-as-you-go

How to apply

For existing VMs, update the licence type via the Azure portal, CLI, or ARM template:

az vm update --resource-group myRG --name myVM --license-type Windows_Server

For SQL databases, apply during creation or update the licence model to “Azure Hybrid Benefit” in the database’s Configure blade.

Tracking

Azure Cost Management shows Hybrid Benefit utilisation. Review it quarterly to ensure you are using all available licences but not over-declaring (which violates the licence terms).

6. Clean up orphaned resources

Every Azure subscription accumulates waste over time. Common orphans:

  • Unattached disks: VMs deleted but disks retained
  • Unused public IP addresses: $3.65/month each (same as AWS EIPs)
  • Empty resource groups: no cost, but clutter makes cost analysis harder
  • Old snapshots: disk snapshots from months ago that nobody needs
  • Unused App Service plans: plans with no apps deployed still incur charges
  • Idle Application Gateways: the base cost runs whether traffic flows or not ($0.246/hour for Standard_v2)
  • Unattached Network Interfaces: left behind after VM deletion

Finding orphans

Azure Advisor catches some of these. For a comprehensive sweep, use Azure Resource Graph queries:

# Find unattached disks
az graph query -q "Resources | where type == 'microsoft.compute/disks' | where properties.diskState == 'Unattached' | project name, resourceGroup, properties.diskSizeGB, sku.name"

# Find unused public IPs
az graph query -q "Resources | where type == 'microsoft.network/publicipaddresses' | where properties.ipConfiguration == '' | project name, resourceGroup"

Schedule these queries as monthly checks. Even small orphans compound: 50 unused public IPs cost $182/month, $2,190/year.

7. Optimise Azure SQL and database costs

Database services are often the second-largest cost after compute. Common waste patterns:

Right-size DTUs or vCores

Azure SQL databases provisioned at S3 (100 DTUs) often run at 10-20 DTU utilisation. Check the DTU utilisation metric in Azure Monitor. If P95 is below 50% of provisioned DTUs, scale down.

For vCore-based pricing, the same principle applies: check CPU and memory utilisation and adjust the tier accordingly.

Use elastic pools

If you run multiple SQL databases for different tenants or microservices, elastic pools share resources across databases. Ten S2 databases at $75/month each ($750 total) might run comfortably in a single 100 eDTU pool at $150/month.

Consider serverless compute

Azure SQL Serverless pauses the database after a configurable idle period and auto-scales compute based on demand. It is ideal for dev/test databases, internal tools, and workloads with unpredictable usage patterns. You pay for compute only when the database is active.

Managed Instance vs SQL on VMs

For SQL Server workloads migrated from on-premises, compare the cost of Azure SQL Managed Instance versus SQL Server on VMs with Hybrid Benefit. The right choice depends on feature requirements, but Managed Instance eliminates OS patching and backup management.

8. Use Azure Spot VMs for fault-tolerant workloads

Azure Spot VMs use spare capacity at up to 90% discount versus On-Demand pricing. The trade-off: Azure can evict the VM with 30 seconds notice when it needs the capacity back.

Good candidates for Spot:

  • Batch processing and data pipelines
  • CI/CD build agents
  • Rendering and transcoding jobs
  • Dev/test environments (combined with auto-shutdown)
  • AKS node pools for non-critical workloads

Not suitable for Spot:

  • Production databases
  • Stateful services without fast failover
  • Real-time APIs with strict latency SLAs

AKS Spot node pools

AKS supports Spot node pools natively. Add a Spot pool alongside your regular node pool and use taints and tolerations to schedule fault-tolerant workloads on Spot nodes:

az aks nodepool add \
  --resource-group myRG \
  --cluster-name myAKS \
  --name spotnodes \
  --priority Spot \
  --eviction-policy Delete \
  --spot-max-price -1 \
  --node-count 3 \
  --min-count 0 \
  --max-count 10 \
  --enable-cluster-autoscaler

9. Set up budget alerts and anomaly detection

Cost surprises are preventable. Azure Cost Management provides two mechanisms:

Budgets

Create budgets at the subscription or resource group level with alert thresholds:

  • 50% of budget: informational alert to the engineering lead
  • 80% of budget: warning to the team channel
  • 100% of budget: alert to finance and engineering management
  • 120% of budget: trigger an Azure Automation runbook to restrict new deployments

Budgets can trigger action groups, which means you can automate responses: send a Slack message, create a PagerDuty incident, or run a script.

Cost anomaly detection

Azure Cost Management’s anomaly detection uses machine learning to identify unusual spending patterns. It catches:

  • A developer spinning up GPU VMs for testing and forgetting to delete them
  • An auto-scaler overreacting to a traffic spike and not scaling back down
  • A misconfigured Log Analytics workspace ingesting terabytes of verbose logs
  • A compromised service principal creating crypto-mining VMs

Enable anomaly alerts in Cost Management > Cost alerts > Anomaly alerts. Configure them to notify your Slack or Teams channel.

10. Enforce tagging and governance with Azure Policy

Cost optimisation is not a one-time project; it is a discipline. Azure Policy enforces guardrails that prevent waste from accumulating:

Required tags

Create a policy that denies resource creation without required tags:

  • team: which team owns the resource (for chargeback)
  • environment: production, staging, dev, test (for shutdown automation)
  • cost-centre: for finance allocation
  • expiry-date: when the resource should be reviewed or deleted

Allowed VM sizes

Restrict VM sizes in non-production subscriptions to prevent developers from provisioning D64s_v5 instances for testing:

{
  "if": {
    "allOf": [
      {
        "field": "type",
        "equals": "Microsoft.Compute/virtualMachines"
      },
      {
        "not": {
          "field": "Microsoft.Compute/virtualMachines/sku.name",
          "in": ["Standard_D2s_v5", "Standard_D4s_v5", "Standard_B2s", "Standard_B2ms"]
        }
      }
    ]
  },
  "then": {
    "effect": "deny"
  }
}

Region restrictions

Limit resource deployment to approved regions to prevent accidental cross-region deployments that increase data transfer costs and complicate compliance.

Inheritance with management groups

Apply policies at the management group level so they cascade to all subscriptions beneath. This ensures new subscriptions inherit cost governance from day one.

Putting it all together

The 10 techniques above are not independent; they work best as a system:

  1. Visibility first: set up Azure Cost Management dashboards and tagging (techniques 9 and 10)
  2. Quick wins: right-size VMs, shut down dev/test, clean up orphans (techniques 1, 3, 6)
  3. Structural savings: Reserved Instances, Hybrid Benefit, storage tiers (techniques 2, 4, 5)
  4. Advanced optimisation: Spot VMs, database tuning, AKS-specific strategies (techniques 7, 8)

For teams that want a structured approach to Azure cost optimisation, from initial assessment to implementation and ongoing governance, our cloud cost optimisation service combines tooling with architectural review. We also recommend pairing this with cloud cost management tools for multi-cloud visibility.

Key takeaways

  • VM right-sizing is the single highest-impact optimisation on most Azure bills
  • Reserved Instances and Savings Plans save 30-72% on committed workloads; start with 70% of your steady-state baseline
  • Shutting down dev/test environments outside business hours saves up to 70% on those resources
  • Storage lifecycle policies automate tier transitions and prevent accumulation of expensive hot-tier data
  • Azure Hybrid Benefit can save 40-80% if your organisation holds Windows Server or SQL Server licences
  • Budget alerts and anomaly detection catch cost surprises before they become end-of-month shocks
  • Azure Policy enforces governance at scale so optimisation gains are not eroded by new deployments

Need help reducing your Azure bill or building a FinOps practice? Book a free consultation and we will review your Azure subscriptions together.

#azure #cost-optimization #finops #aks #reserved-instances

Related service

Need hands-on help? Our certified engineers deliver this as a managed service.

Cloud Cost Optimisation
Found this useful? Share it with your team or link to it from your documentation.

Need help implementing this?

Our certified engineers can handle the setup, migration, or ongoing operations for you.