When your own infrastructure pays off, and when it does not

"The cloud is expensive" and "running your own servers is not worth it" are both too broad. Which option is cheaper comes down to a handful of questions, and to costs that appear on no price list.

What is usually missing from the calculation

Comparisons almost always capture only the monthly rental cost. They should also include:

• Time spent operating it. Updates, monitoring, certificates, incidents. With your own hardware, also replacement and warranty.
• On-call. Who responds when something fails at three in the morning? That question costs money however it is answered.
• Data transfer. With the large providers, outbound traffic is its own line item and quickly the biggest one for data-heavy applications.
• Redundancy. A single server of your own is not comparable to a managed service that absorbs failures itself. A fair comparison needs two servers and something in front of them.

What you actually pay the cloud for

The premium over rented hardware buys essentially three things:

1. Elasticity: capacity in minutes rather than weeks, and gone again when it is not needed.
2. Managed services: database, object storage, queues, without anyone operating them.
3. Transferred responsibility: hardware, data centre and baseline security sit with the provider.

If you need none of the three, you are paying for something you do not use.

When your own or rented servers make sense

There is a strong case when:

• the load is steady: a system under similar load around the clock gains nothing from elasticity;
• there is a lot of outbound traffic;
• data residency is bound to a specific location by contract or regulation;
• somebody who looks after systems is on the team anyway.

When the cloud is the better choice

• The load fluctuates heavily or is unpredictable.
• The project is young and the requirements are still moving.
• There is nobody who can or wants to take on operations.
• You use many managed services whose self-hosted equivalents would be a project of their own.

The most common mistake: an unplanned mix

Cost rarely explodes because of the decision itself, but because of a grown mixture: a few servers here, a few services there, three ways in, no common monitoring. The operational effort adds up while the advantages of both worlds cancel out.

If you mix, mix deliberately, along a line you can explain, for example: data storage in your own data centre, everything short-lived in the cloud.

How to reach a number you can rely on

1. Calculate both options over 36 months, not one month.
2. Include working time, at a realistic hourly rate.
3. Price in the migration, one-off but rarely small.
4. Think about the exit: what would it cost to switch again in two years?

If both options end up close together, the decision is not about price but about who is going to run the system. That question is easier to answer honestly.