Azure Cost Optimisation · Azure
Published
Azure estates tend to arrive rather than get designed: an enterprise agreement, a migration from the data centre, workloads lifted with their licences and habits intact. The waste that follows has a distinctive shape. Where an AWS bill hides its excess in unused capacity, an Azure bill hides much of its excess in unclaimed entitlements and unbounded defaults, and the largest savings often need no engineering at all — just someone who knows the entitlements exist.
Hygiene: what the estate forgot
The fast, safe money looks much like it does on AWS, with Azure-specific traps:
- Deallocated is not deleted. A deallocated VM stops billing compute, but its managed disks and static public IPs bill on. Every migrated estate carries a museum of “temporarily” stopped machines paying rent through their disks.
- Orphaned disks and snapshots. Deleting a VM does not delete its disks by default. Unattached managed disks are the Azure counterpart of unattached EBS volumes, and they are everywhere.
- Non-production running like production. Auto-shutdown schedules for development VMs are built into the platform, and Dev/Test pricing gives non-production subscriptions discounted rates with no Windows licence charge. Many estates use neither.
- Right-sizing against evidence. Azure Monitor shows what a workload actually uses, and Azure Advisor will point at oversized VMs. As with every platform’s advisor, it identifies; it does not decide or act, and telling waste from deliberate headroom still takes judgement.
Licensing: the layer AWS does not have
The distinctive Azure lever is that Microsoft workloads carry Microsoft licences, and Azure will either charge you for them or credit you for them depending on configuration:
- Azure Hybrid Benefit applies existing Windows Server and SQL Server licences with Software Assurance to Azure resources. Combined with reservations, Microsoft quotes up to around 85% off pay-as-you-go for Windows VMs; for SQL Server, where licensing dominates the price, the effect is often larger still.
- The catch is that none of it is automatic. Hybrid Benefit is enabled per resource, and estates that qualify but never configured it are paying for licences they already own. A licensing review against the estate is one of the highest-return afternoons in Azure cost work.
Commitments: reservations and the savings plan
Azure’s commitment instruments mirror AWS’s, with one welcome difference: reservations (up to around 72% off) can be exchanged and, within limits, refunded, which takes much of the fear out of committing. The savings plan for compute (up to around 65%) trades depth for flexibility across families and regions, and Spot VMs run interruption-tolerant work at discounts that can reach 90%. The strategy — optimise first, commit to the baseline, ladder the terms, plan against renewal dates — is platform-independent and covered in our commitment strategy guide. On Azure, sequence the licensing work first as well: Hybrid Benefit changes the rate you should be committing to.
The observability tax
The line item that most often surprises Azure customers is not compute at all. Log Analytics and Microsoft Sentinel bill per gigabyte ingested, diagnostic settings default to shipping everything, and Application Insights will happily record every request at full fidelity forever. The result is a monitoring bill growing faster than the estate it monitors. Scope diagnostic settings to what anyone actually queries, move verbose tables to cheaper table plans, adopt commitment tiers once volumes are predictable, sample application telemetry, and set retention on purpose. Microsoft’s cost management documentation covers the mechanics; the discipline of asking “who reads this?” is the part that saves the money.
Architecture: the layer with the large percentages
As on every platform, the structural savings live in design. Databases on serverless tiers or elastic pools instead of provisioned capacity nobody measures; scale-to-zero platforms such as Functions and Container Apps for spiky workloads; AKS node pools sized and packed against real requests rather than defaults; storage tiered by access pattern, which has a guide of its own. Our largest Azure engagement, a UK public sector organisation running Azure alongside AWS, saved £2.3 million over four years — and the savings came from coordination and continuous attention at least as much as from any single technical change.
That continuous attention is the product: an estate optimised once gives the gain back within a year. Azure estates sit inside our Managed FinOps service on exactly the same terms as AWS and Google Cloud, and the three-day Cloud Cost Health Check will tell you what your estate should cost — licensing included — before you commit to anything. Or just ask us.