Dedicated server, VPS or cloud: choosing by use

“Should we move to the cloud?” The question is often asked as if it were a choice of corporate identity, when it is settled use by use. A dedicated server, a VPS and the cloud are not three levels of modernity: they are three ways of paying for and running the same thing, computing power, with different strengths. The right answers depend on the shape of your workload, your data and the people who will keep it all running.

This article from our Infrastructure and cloud section sets out the differences, the four questions that decide, and what we do ourselves.

What really sets the three apart

All three sell computing power, memory and storage. They differ in what you rent, how you pay, and what remains your responsibility.

• The dedicated server. A whole physical machine, for you alone. Its resources are guaranteed and its price is fixed per month. Changing size means changing machine.
• The VPS (virtual private server). A share of a physical machine, carved out by virtualisation, with reserved resources and a fixed price. Smaller and cheaper than a dedicated server, it is generally easier to resize.
• The cloud. A catalogue of resources and services created and deleted on demand, billed by usage.

What makes the difference is therefore not the technology (many cloud providers also sell VPSs), but elasticity and how you pay. The cloud lets you double capacity for an hour and pay only for that hour. A fixed-price server costs the same whether it runs at 5% or at 95%.

The four questions that decide

For each use, four questions, in this order.

1. 01Is the workload stable or peaky?

   A CRM, an automation tool, an intranet run in much the same way all year round: a fixed price suits them well. A campaign, an application whose audience doubles in an evening, a heavy job once a month benefit from elasticity.

2. 02What does the data contain?

   Health data, personal data in volume, trade secrets: they often decide the location and the jurisdiction before anything else.

3. 03Who will run it?

   A server requires knowing how to maintain a system; the cloud, knowing the console, permissions and billing of its provider. Choose what someone, in-house or at a provider, will still be able to do in two years.

4. 04How much over three years?

   Not the first month’s price: the full cost, operating hours included, compared use by use.

The fourth question is the one that surprises most. The cloud is unbeatable for starting small and absorbing the unexpected; on a stable, known workload, a fixed price often closes the gap. The items that inflate a usage-based bill are detailed in the five items that inflate the cloud bill.

What a well-documented cloud exit shows

The software company 37signals, which runs Basecamp and HEY, moved its applications from AWS to its own machines in 2023, and publishes its figures. On the page it devotes to this exit, it announces savings of about 10 million dollars over five years, a reduction of between half and two-thirds of its infrastructure costs, without hiring anyone extra for operations.

This case says something precise, not a general rule: a company whose workload is stable and known, and which already has people able to run machines, can pay much less outside the cloud. A company whose workload varies a lot, or which has nobody to maintain a system, will not necessarily find the same result. The page’s title says it itself: the cloud is not for everyone, which does not make it a bad choice for the others.

In all three cases, someone runs it

None of the three options removes operating work; they split it differently:

• on a dedicated server or a VPS, you manage the system, updates, security, backups and monitoring; the provider manages the hardware and the network;
• in the cloud, the provider manages more layers, but you still have to configure permissions, networking, backups and the bill, and to understand services that change often.

That is why the choice depends as much on available skills as on technology. A system administrator who knows Linux well will run a server with confidence; a team used to a big provider will be more at ease with its managed services. Both professions remain necessary, and the worst choice is the one nobody will be able to keep up. Monitoring, in particular, never goes away: we detail what it should cover in knowing a service is down before customers do.

What we do ourselves

We combine, and each choice has its reason.

The downside of a single server is well known: if it goes down, everything goes down. We offset it with off-server backups, tested restores and a written recovery plan, which we detail in the disaster recovery plan. Beyond a certain availability requirement, the answer becomes a second server, or an orchestrator, which changes cost and complexity: that is the subject of Kubernetes: when a mid-sized company does not need it.

Combining without multiplying the rules

The risk of a mixed setup is not technical: it is having two ways of managing access, two backup policies and two bills that nobody connects. The answer is a single foundation of rules (named access, tested backups, budgets, documentation), applied wherever your services run. That is what our cloud infrastructure foundation guide describes.

To move a service from one mode to the other without an outage, the method is in cloud migration without breaking things. The question of the provider’s jurisdiction is covered in Europe’s dependence on the American cloud, and upcoming obligations in what NIS2 will require of French companies.

Where to start

List your uses, one per line, and answer the four questions for each. Internal tools with a stable workload, which today cost a per-user subscription, are often the first candidates for a fixed-price server: our guide to which tools to move out of SaaS first suggests an order, the three-year cost calculation the method, and self-hosted n8n in production a concrete example. For whatever stays on a server, do not forget backups and what really breaks.

If you want your business tools on a server of your own, our Leaving SaaS offer installs them in three weeks, with tested backups and documentation included. If your project belongs at a big provider or in a European cloud instead, tell us about it: we will tell you what needs to go in place there.

Frequently asked questions

Sources

1. NIST, SP 800-145, The NIST Definition of Cloud Computing, September 2011 (new tab)csrc.nist.gov
2. 37signals, Cloud Computing Isn’t For Everyone (new tab)basecamp.com