← All articles

On-prem or cloud? Start with the workload, not the price list

"How much would it cost us in AWS?" is one of the most common questions we get. The honest answer is: it depends on what you move, how you move it and what you stop paying for on-prem. A price list answers none of that.

Why quick comparisons mislead

A typical first attempt takes the total number of virtual machines, picks a similar instance size for each one and multiplies by the hourly rate. The result is usually wrong in both directions:

  • Too high, because on-prem VMs are sized for peaks and nobody ever shrinks them. In the cloud you pay for what you allocate, so oversized VMs become an expensive habit.
  • Too low, because storage, backups, data transfer, licenses and support are left out. For many workloads these lines are larger than compute.

How we build the model

Five steps of an on-prem versus cloud cost model: Inventory by workload: owner, purpose, real utilisation from monitoring; Map to a target: lift-and-shift, managed service, or stay on-prem; Count the whole bill: storage, backups, egress, licenses, people; Count what you stop paying: only savings that really end; Show a range: on-demand, committed use, realistic mixHOW WE BUILD AN ON-PREM VS CLOUD COST MODEL01Inventory by workloadowner, purpose, real utilisation from monitoring02Map to a targetlift-and-shift, managed service, or stay on-prem03Count the whole billstorage, backups, egress, licenses, people04Count what you stop payingonly savings that really end05Show a rangeon-demand, committed use, realistic mix
The order matters: a price list applied to an unmapped inventory is where most estimates go wrong.
  1. Inventory by workload, not by host. Each VM gets an owner, a purpose and a real utilisation profile from monitoring, not from the original request.
  2. Map to a target, with options. Lift-and-shift, managed service, or stay on-prem. Each workload gets the option that makes sense for it, not one rule for everything.
  3. Count the whole bill. Compute, storage tiers, snapshots and backups, egress, interconnects, licenses that change model in the cloud, and the people who run it.
  4. Count what you stop paying. Hardware refresh you avoid, data center space and power, support contracts that end. Only savings that really end belong in the model.
  5. Show the range. On-demand, committed use and a realistic mix. Finance trusts a range with assumptions more than a single number.

What usually comes out

The answer is rarely "everything to the cloud" or "nothing to the cloud". Stable, heavy workloads with predictable load often stay cheaper on owned hardware. Spiky, customer-facing and short-lived workloads tend to win in the cloud. The useful result is a list of which workloads move, in which order, and what it does to the budget each year.

If you are facing this decision, write to us. We will tell you what data we need and how long the model takes to build.

Have an infrastructure decision coming up?

Tell us what you are planning and we will come back with how we would approach it.

Email us[email protected]