What is the cloud really?
Summary: the word "cloud" is used for anything and everything. There is however an official and rigorous definition, published by the American NIST, which rests on five mandatory characteristics. This lesson gives that definition, traces the history since 2006, distinguishes public, private and hybrid cloud, and dismantles the misconceptions.
1. The official NIST definition
The NIST (National Institute of Standards and Technology) published in 2011 the definition that remains the worldwide reference: document SP 800-145.
This definition is a practical tool, not an academic curiosity. Use it as a test grid: a hosting provider that bills you a server by the month, with a 24-hour delay for any change, is not selling cloud — it is selling traditional hosting with modernized marketing vocabulary.
The two most discriminating criteria are self-service (criterion 1) and measured service (criterion 5). Many offerings presented as cloud fail on one of the two.
2. The cloud is not "just someone else's server"
This phrase is repeated everywhere. It is technically true but conceptually false, and it is worth understanding why.
The right formulation: the cloud is infrastructure drivable through an API and billed on usage. The hardware's location is just a detail — it is the programmability that constitutes the breakthrough.
3. A short history of the cloud — from 2002 to 2026
The date to absolutely remember: March 2006, the launch of Amazon S3. It is the moment when computing truly became a measured service, like electricity — exactly what McCarthy had imagined in 1960.
Structuring anecdote: AWS was born from an internal Amazon need. Teams were losing weeks waiting for resources, so the company built APIs to free itself. Only afterwards did it realize others would pay for the same service. The worldwide cloud leader is a side effect of an internal reorganization.
4. Public, private, hybrid cloud and multi-cloud
These four terms designate distinct deployment models, often confused.
Important precision about the private cloud: a private cloud keeps NIST criteria 1, 2, 3 and 5, but largely loses criterion 4 — elasticity. You cannot exceed the hardware capacity you bought. It is the fundamental trade-off of the private cloud: you gain control, you lose unlimited elasticity.
Warning about multi-cloud: it is very often adopted for the wrong reasons — "avoiding provider dependence". In practice, it doubles the required skills, complicates security, and is expensive in cross-cloud data transfer. Lesson 5 treats this subject in detail.
5. The five misconceptions to dismiss
Misconception 5 deserves particular attention — it is the most frequent cause of disappointment after a migration. Moving a virtual machine running 24 hours a day as-is to a cloud instance running 24 hours a day produces no savings, and often an extra cost. The gain appears when you rethink the architecture: automatic shutdown outside production, automatic scaling, managed services, storage with differentiated cost tiers.
6. The essential vocabulary
These terms come up constantly. Knowing them keeps you from losing the thread of a technical discussion.
| Term | Meaning |
|---|---|
| Instance | A running virtual machine |
| Provision | Create and make a resource available |
| Elasticity | Ability to adjust resources up and down |
| Vertical scalability | Enlarging a machine (more processors, more memory) |
| Horizontal scalability | Adding identical machines in parallel |
| Auto-scaling | Automatic adjustment of the number of instances based on load |
| Managed service | Service whose provider handles the full operation |
| Region | Geographical area grouping several datacenters |
| Availability zone | Isolated datacenter within a region |
| Egress | Data leaving the cloud — almost always billed |
| Ingress | Data entering the cloud — usually free |
| Vendor lock-in | Difficulty and cost of migrating to another player |
| Reversibility | Ability to retrieve your data and leave |
The most important term in this table for your bill: egress. Getting data into the cloud is free; getting it out is billed, sometimes heavily. It is a deliberate economic mechanism, and one of the main sources of bad surprises. Lesson 6 comes back to it.
7. The market figures in 2026
To know about the European players: their market share remains modest, but their relevance is real and growing for two reasons — sovereignty (European legislation, GDPR, no American extraterritoriality) and price (Hetzner and Scaleway are often significantly cheaper on raw compute).
Remember in 30 seconds
- The NIST defines the cloud by five mandatory characteristics: self-service, broad network access, pooling, rapid elasticity, measured service.
- The most discriminating criteria: self-service and measured service. Many "cloud" offerings fail on them.
- "A server at someone else's place" is reductive: the breakthrough is the API and usage-based billing.
- Birth date of the modern cloud: March 2006, the launch of Amazon S3.
- AWS was born from an internal Amazon need — the worldwide leader is a side effect.
- Four models: public, private (limited elasticity), hybrid, multi-cloud (doubled complexity).
- Most expensive misconception: believing that moving your virtual machines as-is saves money. The gain comes from redesign.
- Crucial term for the bill: egress — outgoing data is billed.
Next: IaaS, PaaS, SaaS and serverless: where your responsibility ends →