Cloud vs On-Premises: What Makes Sense for SMEs

The cloud-versus-on-premises question is usually presented as settled. It is not, and for Swiss SMEs in particular the honest answer is more often “both, deliberately” than “all of one”.

What follows is the comparison without the vendor framing on either side.

Start by separating three things that get conflated

Most arguments go in circles because the participants are discussing different questions:

• Where the hardware is — your rack, a Swiss colocation facility, a hyperscaler region
• Who operates it — your team, a managed provider, the platform itself
• How you pay — capital expenditure with depreciation, or operating expenditure per month

These can be mixed. Managed on-premises exists. So does self-operated cloud that costs more than the rack it replaced. Deciding all three at once with one word is how organisations end up somewhere they did not choose.

Total cost over five years, honestly counted

Comparing a monthly cloud bill to a server purchase is the most common error in this discussion. A five-year view, with everything included, looks different.

On-premises, counted properly:

• Hardware, plus the refresh in year four or five
• Housing or data-centre costs, power and cooling
• Licences, which are often priced differently on-premises
• Operations labour — patching, monitoring, backup verification, hardware failures
• Spare capacity for peak load, paid for year-round
• The cost of downtime at your actual availability, not the aspirational one

Cloud, counted properly:

• Compute and storage, at real usage rather than the estimate
• Data transfer, the line item missing from every projection and present on every invoice
• Managed services, which are more convenient and more expensive than self-operated equivalents
• Operations labour, which shrinks but does not disappear — it changes shape
• The optimisation work that produces the savings, which only happens if budgeted

Where cloud usually wins on cost: variable or unpredictable load, workloads with strong peaks, anything needing geographic reach, and environments that should not exist overnight.

Where on-premises still wins: steady, predictable load on hardware you already own and have not finished depreciating. This case is real and routinely dismissed. Migrating a stable workload off hardware with two years of life left frequently costs more than leaving it.

Operational trade-offs

Cost is the argument people have. Operations is usually what decides it.

Speed of change. Provisioning an environment in the cloud is minutes; on-premises it is a procurement conversation. For a business that needs to try things, this difference outweighs most cost deltas.

Failure modes. On-premises, you own the failure and the fix. In the cloud, you own less of the fix and more of the waiting. Neither is strictly better — but the second requires a different kind of plan, and teams often discover that during the first outage.

Skills. Both need expertise, in different places. On-premises needs hardware, network and storage depth. Cloud needs identity, network architecture and infrastructure-as-code. Assuming your team automatically carries the second because it carried the first is a common and expensive assumption.

Capacity discipline. On-premises, over-provisioning is a one-time purchase you forget. In the cloud it is a monthly charge that compounds — which is why the optimisation step in a cloud migration in five steps is not optional.

Where Swiss regulation actually constrains the choice

This is the part generic comparisons skip, and it is often decisive.

revDSG/nDSG has applied since September 2023. GDPR applies wherever EU personal data is processed. Both require you to know where data sits and who can access it — and these are separate questions. A Swiss region does not answer the access question if support or administration reaches it from another jurisdiction.

FINMA-supervised businesses have additional outsourcing requirements: documented risk assessment, audit rights, and a demonstrable exit path. “We would migrate back” does not satisfy that.

Three practical consequences:

• Data classification comes before the hosting decision. Not all of your data carries the same constraint, and treating it as one block forces the most restrictive answer onto everything.
• Sub-processors are part of your documentation. Your provider has suppliers; that chain is in scope.
• Ask the access question in writing. Residency is in the marketing material. Access control is in the contract, or it is not anywhere.

None of this rules the cloud out. It rules out choosing the cloud without answering these first, then discovering the answer during an audit.

What most Swiss SMEs actually land on

A hybrid split that follows the constraints rather than a philosophy:

• Regulated or highly sensitive data in Switzerland — on-premises, Swiss colocation, or a Swiss cloud region with contractual access limits
• Everything with variable load in the cloud, where elasticity is worth paying for
• Non-production environments in the cloud without exception — they are the clearest win, since they should not exist most of the time
• Stable workloads on unamortised hardware left alone until the refresh date makes the decision for you

That last point is worth repeating because it gets lost in transformation enthusiasm. The hardware refresh date is the natural decision point. Forcing a migration ahead of it usually means paying twice.

A decision sequence that works

1. Classify data by constraint, not by system
2. Establish the real five-year cost of the current setup
3. Identify what has variable load and what does not
4. Check the hardware depreciation schedule
5. Decide per workload, not per company

Step 5 is the one that matters. “We are moving to the cloud” is a slogan; a disposition per workload is a plan.

See also ten cloud migration mistakes mid-market companies still make and our Cloud & DevOps work.

If you want help putting numbers to your own comparison, book a call. No obligation.