Your on-prem TCO is bigger than the invoice — labour isn't the gap

A typical TCO conversation looks like this. We pull out the on-prem invoice: Dell's entry-level rack server now lists around AUD $36,000 capex, with roughly $2,400 a year in warranty and the MSP fee on top. Then the equivalent private cloud monthly bill. The on-prem line items look smaller. The cloud number looks bigger. The customer concludes private cloud is more expensive and the conversation stops.

The two columns aren't the same product. The on-prem line shows the cost of one server with no high availability, no disaster recovery, capacity fixed at whatever was specified two years ago, and a set of physical-infrastructure costs the customer is paying but isn't seeing on any invoice. The private cloud line shows a multi-host cluster with built-in HA, geo-replicated backup, scalable capacity, and the underlying infrastructure absorbed into a single per-month figure. Compared like for like, the columns aren't close.

The infrastructure costs that never appear on the invoice

Start with the costs the customer pays for but doesn't see. A typical SMB server running 24/7 draws around 400 watts under load. At current Victorian small-business electricity rates of roughly 30-40 cents per kWh, that translates to $1,000-$1,400 a year per server in raw power. Cooling adds another twenty to forty per cent on top — every watt of compute power needs a fraction of a watt of cooling to remove it, which is why data-centre PUE benchmarks sit between 1.2 and 1.6 even at scale. UPS battery refresh runs $500-$1,500 every three to five years. Rack space, while small, has a real opportunity cost in floor area.

None of these line items appear on the server invoice or the support contract. They sit on different bills — the electricity account, an electrician's invoice for the next UPS swap, the lease for the office floor that houses the rack. They aren't itemised against the server, which is why most TCO comparisons skip them entirely.

The resilience gap

The on-prem server is a single host. If the motherboard fails, the business stops until parts arrive and the rebuild finishes — that's typically 4-24 hours under a same-business-day warranty, longer if anything in the configuration is non-standard. To match what private cloud provides out of the box — automated failover across multiple physical hosts in the same site — you would buy a second server, shared or replicated storage, and the licensing to cluster them. That doubles or triples the hardware bill. Most SMBs don't do this, so they aren't actually paying the cost of the resilience they think they have.

The disaster recovery gap is worse. To match private cloud's geo-replicated backup, the SMB would need a second physical site or a colocation arrangement, replication links, and the operational discipline to keep the replica current. Most on-prem SMB environments don't have this. Acquiring it on-prem is structurally expensive; in private cloud it sits inside the monthly bill.

The capacity tax

The on-prem box is specified for projected peak load three years out, which means it runs at 20-30% utilisation most of the time. The customer is paying for compute that isn't being used — and paying for it again in power and cooling for the unused capacity. Private cloud charges for what the workload actually consumes, with headroom available on demand. That's the structural utilisation advantage, not a marketing claim.

Where the labour line lives

Here is where the comparison gets honest. If you have a managed service contract, the labour to keep both architectures running is roughly equivalent. We manage on-prem servers for clients and we manage private cloud workloads for clients, and the engineer hours are about the same. The MSP fee doesn't change between the two models.

That is worth saying clearly because it is where most cloud-versus-on-prem pitches lose credibility — they imply you will save on labour. Under an MSP relationship in place, you generally won't. The delta lives elsewhere.

How to actually run the comparison

For SMBs comparing the two options, the right line items to put side by side aren't hardware quote against monthly cloud bill. They are: capex outlay against monthly fee; physical-infrastructure costs against zero; cost of acquiring HA and DR on-prem against HA and DR included; depreciation risk on an asset whose replacement cost is climbing materially against a fixed monthly cost. Labour cancels.

The strategic point is that the on-prem invoice was never meant to reflect TCO. It reflected hardware acquisition and a support contract. That was enough information when memory got cheaper every year, refreshes were predictable, and resilience was a separate conversation most SMBs put off. None of that holds now.

Communicat runs both on-prem and private cloud environments for Victorian SMBs. If your last TCO conversation stopped at the hardware quote and the monthly cloud bill, the conversation worth having is the one above — full stack against full stack, with the resilience and infrastructure deltas surfaced. That comparison usually flips the answer.