Own Server or Cloud: An Architecture Decision, Not a Belief System

In many mid-sized businesses, the core IT infrastructure runs on a physical server sitting somewhere in the building. In the boiler room, in the archive, sometimes even in the attic. It worked well for a long time, until it didn’t.

The temptation to simply buy the next server is strong. It is usually the wrong move.

What the own server really costs

The purchase is only the tip. What actually accrues:

• Hardware: replaced before it becomes a risk
• Licences: server operating system, mail server, backup software, annually
• Maintenance: updates, patches, reboots – someone has to handle it
• Power and cooling: a server room needs cooling, often around the clock
• Failure risk: what happens when the server is down because a fan dies
• Security compliance: backup strategies, disaster recovery, access controls

Cloud providers have solved most of these points better – not because they are smarter, but because they optimise across thousands of customers simultaneously. What is a special investment for an own server is standard equipment for a cloud provider.

When cloud migration makes sense

It makes sense when at least one of the following conditions applies:

• Hardware is due for refresh. Anyone who has to buy new hardware anyway should first ask whether they want to buy new hardware at all.
• Employees work increasingly mobile. VPN connections to an own server are a crutch that becomes more expensive as the number of remote workers grows.
• Multiple locations share data. An own server is a local construct; once multiple locations are involved, the cloud becomes structurally superior.
• Email causes repeated problems. Own mail servers are the most common source of IT tickets in mid-market businesses. Cloud mail (Microsoft 365 or Google Workspace) resolves the majority of these problems structurally, not symptomatically.
• Backup strategy is unclear. Anyone who cannot explain at any given moment where their data is backed up, how long it is retained, and how quickly it can be restored – cloud migration is a good opportunity to get this up to standard.

When it does not make sense

It does not make sense when:

• The industry or activity does not require it. Some workshops, some production environments, some consulting firms operate perfectly well with local hardware and need no scaling. If there is no pain, no migration needs to start.
• Internet connectivity is unreliable. Anyone in a rural location with unstable connectivity should solve the connectivity first before moving critical systems to the cloud. Cloud mail that regularly goes offline is worse than on-premise mail that works.
• Special compliance requirements exist that can only be met on-premise. This is rare in the SME space, but occurs in certain regulated industries and some public procurement contexts.

The three questions before the decision

Before anyone says “cloud migration”, three questions should be answered:

1. What gets migrated first. Email, file storage, telephony, accounting. Sequential, not parallel.
2. Who conducts the migration. Internal staff rarely have relevant experience. Service providers are necessary, but they need someone on the client side who maintains oversight.
3. What stays on-premise. A complete cloud migration is rarely realistic. Which systems remain local, and how they connect to the cloud, is the most interesting question.

An own server is not the same as own infrastructure

An apparent contradiction that comes up regularly: here I advise against the server in the boiler room, elsewhere I recommend running AI on your own infrastructure so company data stays in-house. That is not a contradiction, it is the same standard applied twice. The question is never “own or external”, but: who operates more reliably, and who controls the data.

The old server in the archive loses on both axes. It is neither professionally operated nor particularly protected, just familiar. Standard workloads like mail and file storage therefore belong with a specialised provider. Confidential company knowledge that an AI is supposed to process, however, belongs under your own control: on dedicated, professionally hosted infrastructure assigned to your business.

The biggest misconception

“Cloud” does not mean “we outsource and no longer need to think about it”. Cloud relocates the responsibility, it does not eliminate it. A company that ran a poor architecture on its own server will run a poor architecture in the cloud. The homework is the same – it just runs somewhere else.

The advantage is not less responsibility. The advantage is that responsibility shifts to what you should actually be doing: thinking about architecture instead of replacing fans.