Azure to AWS: a 19-workload estate assessed with the same rigour as a 700-instance one

The challenge

This was a cloud-to-cloud assessment rather than a data centre exit, and it presented a different problem from every other engagement in the portfolio.

The estate was mostly platform services rather than virtual machines: Azure VMs, AKS node pools, App Service plans, PaaS SQL and PaaS PostgreSQL. There was no conventional server inventory to right-size and no utilisation agent reporting from a hypervisor. Every source object had to be mapped from a platform construct to a comparable AWS target before any pricing was possible, and each mapping had to be defensible or the resulting number would be meaningless.

There was a commercial temptation too. Estates this size are routinely handed a rule-of-thumb estimate. At $30,000 to $50,000 a year, the absolute savings look modest against the effort of a full assessment. But the business still needs to know whether the move is commercially sound, and a rule of thumb cannot answer that question with any authority.

Our approach

We published an explicit mapping table covering every source object, with the assumptions recorded alongside each one:

We flagged two mappings as deliberately conservative rather than optimal. AKS maps naturally to Amazon EKS, but at one node per cluster we advised EC2 with container tooling as the lower-cost target, and noted EKS as the correct answer once node counts reach three or more. PostgreSQL was modelled on EC2, with Amazon RDS for PostgreSQL identified as the true like-for-like target that would replace the raw EC2 line under a Multi-AZ production configuration.

We advised against EC2 Dedicated Hosts on Windows licensing. The estate was predominantly Windows Server 2025, which would require a downgrade to make mobility available, and at this scale the cost of running Dedicated Hosts would exceed any licence saving available.

We modelled the estate as a direct match as well as right-sized, so the business could see what the analysis was worth in isolation.

The results

The recommended position sits 42% below the committed direct-match figure and 55% below direct match on demand, and requires 8 SQL Server Standard cores with active Software Assurance.

SQL Server consolidation reduced the forward requirement from 10 Standard cores to 6, releasing four cores against future renewals. Disaster recovery across all 19 servers priced at $16,418 a year, and the single virtual desktop on Amazon WorkSpaces at $1,874.

Why it matters

What the business received was not a headline number they had to trust, but a mapping table they could challenge line by line, including two places where we had deliberately not chosen the AWS-native option because it would have cost them more at their current scale. That is the difference between a cost estimate and a business case.