Azure pricing for managed disks follows a straightforward model: higher performance tier costs more. Premium SSD costs more than Standard SSD. Ultra Disk costs more than Premium. The assumption baked into most purchasing decisions is that you're paying for performance you'll actually use.
That assumption breaks down as soon as you look at VM IO limits.
The ceiling you're not checking
Every Azure VM has a maximum uncached disk throughput — the hard limit on how much data can move between the VM and its attached disks, regardless of what those disks are capable of.
B-series VMs cap between 22 and 35 MB/s. Some D-series configurations cap at 192 MB/s. Standard Premium SSD P30 delivers 200 MB/s. Standard Premium SSD P20 delivers 150 MB/s.
When the VM ceiling is lower than the disk's throughput capability, the disk is throttled to the VM limit. Performance is identical to a lower-tier disk. Cost isn't.
The pattern in practice
In most Azure estates I look at, disk SKU decisions were made at provisioning time based on the workload type rather than the VM family being used. Premium SSD for anything database-adjacent, Standard SSD for everything else. That heuristic ignores the VM IO ceiling entirely.
The result: a significant fraction of Premium SSD disks attached to VMs that can't use their full throughput. The disk delivers 35 MB/s. It's capable of 150 MB/s. The price difference is £46/month on a 512 GB disk.
Scale that to 50 affected disks across an estate and the annual overspend is around £27,600. In one review, disk SKU alignment alone accounted for £30,000 per annum in recoverable waste.
Where the problem concentrates
B-series VMs are the clearest case — the IO ceiling is low enough that most data disk sizes exceed it. But the pattern appears elsewhere:
General-purpose VMs running management or monitoring workloads where disk IO is never the constraint. DevOps agent pools where the bottleneck is network or CPU, not storage. Dev/test environments where production-tier disks were copied without reviewing whether the performance tier made sense for the environment.
The common thread: disk SKU inherited from a template or copied from production without evaluating whether the VM family can actually use the performance tier being paid for.
What changes when you find it
Changing disk SKU from Premium to Standard SSD on a stopped VM is a non-destructive operation. The data is preserved. The capacity is unchanged. The throughput delivered to the VM is identical if the VM IO ceiling was already below the Premium SSD's capability.
It's one of the lower-risk cost changes in Azure, which is why it tends to appear as a quick win in structured reviews.
Disk and VM SKU alignment is part of our standard Azure cost review. If you'd like a read-only look at your estate, the free cost assessment identifies this pattern alongside others.