What is Cloud Computing?
Reading time: 15 min
Article updated in 2026.
Cloud computing has established itself as one of the main pillars of digital transformation . It hosts business applications, customer data, collaborative tools, marketing platforms and, increasingly, artificial intelligence services.
Yet it is often reduced to an overly simplistic idea: that of files or software stored "somewhere on the Internet"... The cloud does not simply refer to a remote hosting location . It corresponds to a model for the provision and consumption of computing resources, in which computing, storage, network, or processing capacities can be made available on demand.
For businesses, the question is no longer simply whether to use the cloud. It now consists of determining which services to choose, for which uses, at what cost, and under what governance rules.
What is cloud computing?
The National Institute of Standards and Technology, better known as NIST, defines cloud computing as a model that allows access, on demand and over the network, to a shared set of configurable computing resources.
These resources can include networks, servers, storage space, applications, or services. They can be quickly made available, adjusted, or withdrawn without requiring significant manual intervention from the provider.
This definition reveals five essential characteristics.
- The service must first and foremost be available on demand. A user or application can obtain a resource without having to systematically request human intervention.
- It must then remain accessible via the network, from different types of terminals or environments.
- Resources are generally shared between several clients, although the data and processing of each remain logically separate.
- The cloud is also characterized by its elasticity. Capacity can increase or decrease depending on the level of activity.
- Finally, consumption is measured. The provider can account for storage volume, computing power, traffic, number of requests, or other units of use.
"The cloud is not a place: it's a way of providing, operating, and consuming computing resources."
Pedagogical formulation of the writing.
How does the cloud work?
Behind a browser-accessible interface lies a physical infrastructure comprised of data centers, servers, network equipment, and storage systems.
Virtualization allows the capacity of a single physical server to be divided into several independent environments. Containers facilitate the deployment of applications in a lightweight and portable form. Automation and orchestration tools then distribute resources, monitor their status, and adapt their capacity.
Application programming interfaces, or APIs, also play a central role. They allow applications to communicate with each other, trigger processes, and exchange data.
The general operation can be represented as follows:
Cloud computing
How does the cloud work?
A request passes through several technical layers before mobilizing the necessary resources.
1
User or application
A person, a piece of software, or a terminal makes a request.
↓
2
Internet or private network
The request is transmitted via the Internet, a VPN, or a secure network.
↓
3
Cloud service
The platform receives the request and controls access rights.
↓
4
Shared and orchestrated resources
The capacities are distributed between servers, networks, storage and automated services.
↓
5
Available services and capabilities
To remember :
The cloud allows the use of computing resources without directly managing all the underlying infrastructure.
The user does not necessarily know the physical server that executes their processing. They consume a service whose technical complexity is largely handled by the provider.
This abstraction is one of the main advantages of the cloud. It allows companies to focus on their applications and their use cases rather than on the day-to-day management of each hardware component.
SaaS, PaaS and IaaS: three levels of service
Cloud computing is not a single service. It encompasses several levels of service, which grant varying degrees of control to the client company.
SaaS: using a ready-to-use application
Software as a Service, or SaaS , refers to an application accessible online, usually by subscription.
The publisher manages the infrastructure, servers, updates, and most maintenance. The client configures the application, manages its users, and uses its data.
In the Martech field , a CRM , an email platform , a marketing automation tool, a content management solution or an analytics service can be offered in SaaS format.
Example: A marketing team uses an online platform to create campaigns, manage contact lists, and analyze performance. It does not administer the servers that run the application.
PaaS: Building and Deploying Applications
Platform as a Service, or PaaS, provides an environment that allows for the development, testing, and deployment of applications.
The provider manages the infrastructure, the operating system, and some of the technical components. This allows developers to focus on the code, data, and business logic.
Example: A company develops an application that calculates propensity scores from its CRM data. It uses a cloud platform to run the code, expose an API, and automatically manage load increases.
IaaS: Accessing infrastructure resources
Infrastructure as a Service, or IaaS , provides virtual machines, networks, storage capacity, and other technical resources.
The supplier manages the physical hardware. The company retains more control over systems, applications, and configurations. It also assumes a larger share of administration and security responsibilities.
Example: an organization deploys a customized marketing data platform on virtual machines and storage spaces that it configures itself.
The sharing of responsibilities can be summarized as follows:
SaaS: use
The application is largely managed by the provider.
PaaS: building
The company develops its services on a managed platform.
IaaS: Configure and administer more
The company controls a larger part of the technical environment.
None of these models is inherently superior to the others. The right choice depends on the desired level of customization, available skills, security requirements, and the expected degree of control.
Public, private, hybrid and multicloud
Cloud services can also be classified according to their deployment method.
The public cloud
In a public cloud, resources are offered by a provider to multiple organizations. Infrastructure is shared, even if customer environments and data are separate.
This model facilitates access to significant capacity and a vast catalog of services. It generally offers high elasticity and rapid deployment.
The term "public" does not mean that the data becomes public. It indicates that the infrastructure is offered to multiple clients according to a shared services model.
The hybrid cloud
Hybrid cloud combines internal resources, a private cloud, and public cloud services.
A company can, for example, keep certain databases in its information system while using cloud services for analytics, marketing campaigns or artificial intelligence.
This approach often responds to historical, regulatory, or operational constraints. However, it increases the complexity of integrations and monitoring.
The private cloud
A private cloud is reserved for a single organization. It can be operated in its own data centers or hosted by a provider.
This model offers greater control over architecture and security policies. However, it requires more investment, skills, and administrative capacity.
Multicloud
A multicloud strategy relies on the use of multiple cloud service providers.
It can help to select the service best suited to each need, to respond to geographical constraints or to reduce certain forms of dependence.
However, using multiple clouds does not automatically guarantee independence. An application may remain heavily dependent on APIs, databases, or mechanisms specific to a particular provider.
Multicloud also multiplies the environments to secure, the contracts to monitor, the skills to maintain, and the invoices to analyze.
Why has the cloud become essential to Martech?
Most modern marketing technologies rely directly or indirectly on cloud services.
CRM systems centralize sales information and customer interactions. Customer Data Platforms ( CDPs ) aggregate data from various sources. Marketing automation platforms orchestrate campaigns. Digital Asset Management (DAM) tools manage images, videos, and documents. Analytics services analyze customer journeys and conversions.
The cloud facilitates connections between these different tools via APIs. It also allows processing capacity to be adapted to data volumes and activity peaks.
Let's take the example of a team preparing a personalized campaign.
The website forms, CRM, mobile app, and customer service collect information with the necessary consents. A CDP consolidates this data and matches profiles associated with the same individual.
The team then creates a customer segment likely to renew their subscription. An artificial intelligence service suggests several content variations. An orchestration platform selects the appropriate channel and triggers the delivery.
The interactions produced by the campaign finally return to the data environment in order to measure results and improve subsequent actions.
The journey can be summarized as follows:
Martech Path
From data to campaign measurement
Data flows between several components before being activated, analyzed, and reconciled.
1
Collection of data
Websites, applications, points of sale and customer service
↓
2
Centralization of profiles
CRM and CDP
↓
3
Analysis and decision
Segmentation, scoring and artificial intelligence
↓
4
Campaign orchestration
Defining scenarios, rules, timing, and activation channel
↓
5
Omnichannel activation
Email, web, advertising, social media and customer service
↓
6
Measure and reconciliation
Attribution of results, performance analysis, and data feedback within the Martech ecosystem
A continuous journey:
The campaign results feed back into the CRM, CDP and analytics models to improve future activations.
In this scenario, no single platform necessarily works alone. The value comes from the ability to flow data between multiple services within a controlled framework.
The advantages of the cloud, without the misconceptions.
The cloud allows for the rapid deployment of new services. A team can create a test environment, open up storage capacity, or activate a tool without waiting for the installation of new physical servers.
Its elasticity helps businesses absorb fluctuations in activity. A platform can mobilize more resources during a major campaign, then reduce its capacity once the peak has passed.
The cloud also provides access to advanced services: managed databases, analytics tools, image recognition, machine translation, machine learning models or generative AI.
It facilitates collaboration between teams spread across several regions and accelerates certain experiments.
However, these advantages should not lead to the idea that the cloud is always simpler, cheaper, or more secure.
The cost depends on consumption, architecture, contracts, and management discipline. A poorly designed application can generate significant storage or transfer volumes. Test environments may remain active even when not in use. Redundant services may be purchased by multiple teams.
In a traditional infrastructure, a company often commits a significant portion of its expenses before use: hardware, licenses, or the construction of a data center. These investments primarily fall under capital expenditures, or Capex.
In the cloud, costs are more often in the form of operating expenses, or OpEx. They vary with consumption. This flexibility can facilitate projects, but it also makes bills less predictable.
The costs and role of FinOps
FinOps is a discipline that brings together technical, financial, and business teams to improve the value obtained from cloud spending.
It's not just about reducing bills. It's about knowing who consumes what, for which project, with what result, and at what level of service.
A FinOps approach may include budgets, alerts, cost allocation by team, sizing rules and regular reviews of unused resources.
It also helps to arbitrate between several options: buying reserved capacity, paying on a usage basis, using a managed service or developing a solution in-house.
Artificial intelligence further complicates this analysis. Costs can depend on the number of tokens processed, the model used, the GPU usage time, the volume of vector databases, API calls, and data flows.
A poorly configured autonomous agent can make multiple calls to a model or repeat a process without producing additional value. Automation therefore does not eliminate the need to control usage; on the contrary, it can accentuate it.
Security: a shared responsibility
Moving to the cloud does not transfer all security responsibility to the provider.
This typically protects data centers, physical equipment, networks, and certain software layers. The precise allocation depends on the service chosen.
The client company remains responsible for many elements: account management, access rights, configurations, imported data, terminals, passwords, API keys and authorized uses.
With a SaaS application, the provider assumes much of the technical operation. However, the client must still properly define roles, manage employee accounts, and control shared data.
With an IaaS infrastructure, the company manages more components. Its scope of responsibility therefore becomes broader.
Example: misconfigured storage
A storage space can rely on a robust and redundant infrastructure. If it is configured to allow uncontrolled public access, the data remains exposed. In this case, the problem does not necessarily stem from the physical security of the provider, but from the configuration applied by the client.
Cloud security must therefore combine the guarantees of the provider and the quality of internal governance.
Dependence, reversibility and interoperability
The cloud can create a technical or economic dependence on a provider. This situation is often called vendor lock-in.
It appears when an application relies heavily on proprietary formats , specific databases, exclusive functions, or APIs that are difficult to replace.
Migration is no longer simply about moving data. It may become necessary to rewrite part of the application, train teams, and rebuild operating mechanisms.
Data egress costs can also factor into the calculation. An inexpensive service for storing information can become more expensive when the company wants to transfer it to another platform.
Therefore, reversibility must be considered before signing the contract. Export formats, deadlines, costs, support offered, and data recovery conditions should all be verified.
The use of open standards, the documentation of architectures and the restoration tests limit certain risks.
An exit strategy does not imply that a departure is planned. It allows one to retain decision-making power.
Data sovereignty and localization
The location of a data center is not sufficient to determine the level of sovereignty of a cloud service.
At least three dimensions must be distinguished:
- the physical location where the data is stored;
- the jurisdiction to which the supplier or its parent company is subject;
- operational control exercised over the infrastructure, access, and encryption keys.
Data hosted within the European Union may be processed by a company subject to non-EU legislation. Depending on the context, this situation can create a risk of access by foreign authorities.
However, not all data has the same level of sensitivity. The requirements for a commercial newsletter are not necessarily the same as those for a health database, information relating to minors, or data sensitive to the state.
In France, the SecNumCloud qualification provides a security and protection framework adapted to certain sensitive uses. However, it is not a general requirement for every cloud project.
The choice must take into account the nature of the data, the risks, the sector of activity, regulatory obligations and the potential consequences of unauthorized access.
The cloud in the age of artificial intelligence
The rise of artificial intelligence strengthens the role of the cloud.
Companies can access models via an API, rent GPU capacity, deploy vector databases, or build search-augmented generation systems, often referred to by the acronym RAG.
The RAG (Research Access Group) allows a model to search for information within an organization's documents before generating a response. This approach can improve relevance and limit certain discrepancies, provided that the sources are well-selected and access rights are respected.
The cloud also facilitates the deployment of agents capable of chaining several actions: querying a database, calling an API, analyzing a document, creating a task or triggering a campaign.
This accessibility accelerates experimentation. A company does not need to build a computing center to test an AI feature.
However, it must examine several risks: confidentiality of data sent to the model, retention of queries, traceability of results, access control, resource consumption and dependence on proprietary technology.
In a martech environment, the use of AI must also respect the purposes defined for customer data and the rules of consent.
The cloud makes AI more accessible, but it does not eliminate the need for governance or evaluation of the value produced.
What environmental impact?
Cloud terminology can give the impression of an immaterial infrastructure. In reality, each service relies on data centers, servers, networks, cooling systems, electricity, water, and physical equipment.
Sharing resources can improve infrastructure utilization. A large provider can distribute the workload across multiple customers and optimize certain equipment more efficiently than a single organization.
This efficiency, however, does not guarantee an automatic reduction in overall impact. The increasing number of uses, the growing volume of data, and the development of artificial intelligence can increase the demand for electricity and resources.
The impact depends in particular on the energy mix, the efficiency of data centers, the lifespan of the equipment, the level of server usage and the amount of data transferred.
GreenOps applies a logic of measurement and optimization to the environmental consequences of digital technology. It can lead to the elimination of unused resources, adapting power to needs, limiting the retention of valueless data, and choosing hosting regions according to documented criteria.
Like FinOps, GreenOps requires making energy consumption visible. What is neither measured nor attributed remains difficult to improve.
How to choose a cloud solution?
Choosing a cloud solution should not begin with comparing technical catalogs. It should start with the business need.
A campaign platform, a customer database, and a model training environment do not meet the same requirements.
The first question therefore concerns the expected value. What problem is the service supposed to solve? Which users will benefit from it? How will its effectiveness be measured?
The company must then examine the integration with its information system. Does the service have documented APIs? Can it exchange data with the CRM, CDP, analytics tools, and consent management systems?
Security and compliance considerations must be addressed from the design stage. They should not be added after deployment.
The full cost must include the subscription or consumption, but also integration, training, administration, support, data transfers and any possible future migration.
A short decision grid can help structure the analysis:
- What business need does the service cover?
- What categories of data will be processed?
- Which systems will it need to communicate with?
- What level of availability is actually necessary?
- How will identities and rights be managed?
- Can the data be exported to a format usable?
- What will the total cost be over several years?
- What indicators will allow us to assess the value and impact of the service?
Choosing a cloud solution is therefore not just a technical decision. It involves business units, finance, security teams, legal counsel, data managers, and users.
Conclusion
Cloud computing is no longer an isolated innovation. It has become an essential infrastructure for the operation of businesses and the majority of Martech ecosystems.
It facilitates access to applications, computing power, data tools, and artificial intelligence services. It enables faster project deployment and resource adaptation to needs.
This flexibility comes with new responsibilities. Companies must monitor costs, protect data, manage access, plan for reversibility, and understand the legal constraints that apply to their suppliers.
Cloud maturity, therefore, is not about systematically moving all applications to a remote infrastructure. It relies on the ability to choose the right services, integrate them correctly, and govern them with clear rules.
As artificial intelligence and automation play a more prominent role in marketing platforms, this governance becomes even more strategic. The challenge will not simply be to have more computing power, but to use it in a useful, controlled, secure, and sustainable way.
Some references
- “The NIST Definition of Cloud Computing,” Peter Mell and Tim Grance, National Institute of Standards and Technology, September 2011.
- “Cloud: the risks of a European certification allowing foreign authorities access to sensitive data”, National Commission for Information Technology and Civil Liberties, CNIL, July 19, 2024.
- “FinOps for AI: Tools & Services Considerations,” FinOps Foundation, 2026.
- “Electricity 2024 – Analysis and forecast to 2026”, International Energy Agency, IEA, January 2024.
Read next
