Skip to content

How to work out the real cost of leaving the cloud

How to compare a public cloud bill with owned hardware in a colocation data center: what each side must include, where the cost sits, and how to decide.

Start from invoices, not list prices

A cloud comparison that starts from a pricing calculator is a guess. The bill is the fact. Every public cloud provides a usage export that lists each resource, its hours, its storage, and its data transfer for the month. Twelve months of those exports, plus the invoices, are the input. They show what is actually running, how much of it is idle, which storage tiers are in use, and how much data leaves the region each month, which is usually the surprise.

The export also shows what the calculator never would: reserved instances bought and not used, snapshots nobody deleted, a second region kept warm for a failover that was never tested. Removing them saves money whether the workloads stay in the cloud or move, so count them in the comparison.

Sort every workload into three piles

Not everything should move, and a comparison that pretends otherwise will be wrong in both directions. Each workload goes into one of three piles.

  • Steady: runs all day, every day, with a load that changes slowly. Virtual machines for line-of-business applications, databases, file services, domain controllers, virtual desktops for a stable headcount. This pile is the candidate list, and it is usually most of the bill.
  • Bursty: scales by the hour or runs in bursts. Batch jobs, test environments spun up and torn down, seasonal peaks. These are what the cloud is for, and they stay unless the burst turns out to be permanent.
  • Leave alone: software consumed as a service, global edge services, and anything whose cloud-only managed features the business depends on. Microsoft 365 is not infrastructure and does not move.

The sorting is done from the usage export, not from memory. A workload with a flat usage line for twelve months is steady whatever anyone thinks of it.

Price the cloud side in full

The cloud side of the comparison is the steady pile at what it actually costs, which is more than its compute line.

  • Compute hours at the rate paid, including the idle hours. A server that is busy for eight hours a day is billed for twenty-four.
  • Storage by the gigabyte-month, by tier, including snapshots and the backup copies the provider charges for separately.
  • Egress and inter-region transfer: every byte read from outside the region, restored, or replicated.
  • Managed services on top of the compute: database tiers, load balancers, VPN gateways, firewalls, NAT, monitoring.
  • Backup as a service, which is often billed per protected instance and per gigabyte stored.
  • The support plan, and any licensing the provider bundles into the hourly rate.

In one recent cloud quote we re-priced, the backup lines alone were a fifth of the monthly total, and backup plus networking, which includes outbound and inter-region transfer, came to almost half. Those lines are the most sensitive to assumptions, and the ones a list-price calculator leaves out.

Price the owned side in full

The owned side is a priced colocation footprint and a hardware list, and it has to carry everything the cloud side carries.

  • The cabinet or partial rack, with A and B power, at the facility's monthly rate.
  • Metered power at the facility rate, from a measured or well-estimated draw, with headroom.
  • Transit from two carriers, the cross-connects, and the private WAN back to the offices and to whatever stays in the cloud.
  • Hardware over its life: servers, storage, switches, firewalls, optics and cabling, with spares, and with the warranty terms stated. Refurbished hardware with one year of parts support costs less to buy and leaves more of the risk of a failure with you.
  • Licensing that the cloud bundled and the owned side must buy: Windows Server, SQL Server, remote desktop access, the hypervisor's support contract.
  • Backup: the second copy, where it lives, and the circuit it travels over.
  • Management: patching, monitoring and backups. Either staff time or a managed service, but never zero. Remote hands around the clock is included in the colocation fee.

Compare three ways of paying for the hardware

How the hardware is paid for changes the cash flow more than the total, so price the hardware three ways: bought outright and depreciated over three to five years; supplied inside a monthly fee, so the move is operating expense like the bill it replaces; or a mix, with the long-lived core owned and the parts likely to change rented.

A finance team can then read the comparison the way they read the cloud bill, month by month, and see the payback month on each line rather than a single number at the bottom.

What the comparison must show

A comparison is only as good as the ability to check it. Ours shows every line on both sides with its figure and the assumption behind it: the hours, the rate, the depreciation period, the power draw, the egress volume. Anyone can take the cloud side back to the invoices and the owned side back to the quotes.

It also lists what it excludes, because every comparison excludes something: tax, staff time if the customer keeps management in house, the escalator on the colocation contract, hardware refresh beyond the period, and the workloads in the other two piles.

Decide on cash flow as well as total cost

Total cost over five years is the headline, but three other numbers decide whether a move happens: the payback month, the cash required before it, and the monthly amount afterwards. A move with a large saving over five years and a two-year payback funded from capital is a different decision from the same saving funded monthly, and both are worth seeing.

The last factor is risk, which is better described than priced: what the hardware warranty covers, what a failed component costs in time, what a cloud price change would cost, and what it would cost to replace any instance types the provider has announced it will retire within the period.

The mistakes that undo a comparison

  • Comparing the cloud compute line with the colocation cabinet line and calling it a result.
  • Sizing the owned hardware to the cloud instance sizes rather than to the measured utilization. Cloud instances are often several times larger than the load on them.
  • Forgetting licensing on the owned side, or counting it on the cloud side when it is already bundled.
  • Leaving egress and backup out because they are hard to read from the invoice.
  • Pricing new hardware with a full warranty on one side and refurbished hardware with none on the other without saying so.
  • Moving the bursty workloads. They cost more on owned hardware than in the cloud.

Telnetrix builds this comparison from your invoices as the first step of a move, then designs the cabinet and the network, moves the workloads with a rollback ready, and runs the environment afterwards.

Contact Telnetrix

If this reads like your situation, tell us what you run and where. We reply by email or phone, whichever you prefer.

Opens your email app, addressed to sales@telnetrix.com.