AWS vs Azure: how to choose your cloud (and where Google Cloud fits)

“AWS or Azure?” is one of the first questions asked at the start of a cloud project, and often one of the least useful as phrased. Comparison charts listing hundreds of services rarely help anyone decide: both platforms can host a web application, a database, a Kubernetes cluster or an AI assistant. What makes the difference is where you start from: your tools, your licences, your people, your data and your regulatory obligations.

This article sets out an eight-point decision framework, a comparison table and the place of Google Cloud in the decision, with a European and French angle where it matters (data residency, health data, sovereignty). At Feekr we work on both AWS and Azure and have no commercial preference: our aim is a decision that is documented and owned. For an overview of what we do, see our AWS and Azure cloud consulting page.

AWS, Azure, Google Cloud: what really sets them apart

Comparable foundations

The three hyperscalers share the same building blocks: virtual machines, object and block storage, virtual networks, managed databases, managed Kubernetes, serverless functions, and tools for monitoring, security and cost management. The names differ (EC2, Azure Virtual Machines, Compute Engine; EKS, AKS, GKE) but the concepts are very close.

The real differences lie elsewhere:

• Ecosystem. Azure is the natural extension of a Microsoft environment: Entra ID, Microsoft 365, Windows Server, SQL Server, Power Platform.
• History and breadth. AWS opened the market and offers a very broad catalogue, backed by a large pool of skills and documentation.
• Data and analytics. Google Cloud is often chosen for BigQuery and its data services, and for its integration with Gemini models.
• Organisational model. AWS structures environments as accounts grouped in an organisation; Azure uses subscriptions under management groups, tied to an Entra ID tenant. These models shape how you handle permissions and billing.

What market share tells you (and what it doesn’t)

According to Synergy Research Group, in the second quarter of 2026 Amazon held 28% of the worldwide cloud infrastructure services market, Microsoft 20% and Google 15%. Those figures confirm that all three are solid, long-term players. They say nothing about your situation: a bigger provider is not automatically a better fit for a mid-sized company whose identities live in Entra ID, and a smaller one is not a risk if it covers your needs.

What about Google Cloud?

Google Cloud has a region in Paris (europe-west9, three zones) and a complete catalogue. It deserves a look if data and analytics are at the heart of your project, if your teams already use Google Workspace, or if you need the French sovereign offering run by S3NS (see below). In practice, most organisations we meet are choosing between AWS and Azure, so this article focuses on those two and flags Google Cloud where it changes the picture.

The decision framework: eight criteria

1. Your existing Microsoft estate

This is often the most decisive factor. If you run Microsoft 365, manage identities in Entra ID and operate Windows Server and SQL Server, Azure offers real continuity: the same admin tools, the same support relationship, the same contracts.

The most tangible argument is financial. Azure Hybrid Benefit lets you bring SQL Server licences covered by Software Assurance to Azure SQL Database and SQL Managed Instance; Microsoft says this can save up to 30% or more on those services. Conversely, since 1 October 2022, Microsoft’s licensing terms single out “Listed Providers” (Alibaba, Amazon, Google and Microsoft): the flexible virtualisation benefit introduced on that date does not apply to them, and Microsoft points customers towards buying licences directly from the provider. Windows Server and SQL Server licensing therefore costs out differently depending on the cloud. Run those numbers with your licence agreement in front of you, before any other comparison.

A Microsoft estate does not force you onto Azure, though. Many organisations keep Microsoft 365 for productivity and run their applications on AWS: it is a common, well-supported combination.

2. Identity: Entra ID does not mean you must choose Azure

This is a common misconception. AWS IAM Identity Center supports SAML 2.0 federation with Microsoft Entra ID and automatic provisioning of users and groups via SCIM 2.0. Your staff sign in to AWS with their usual Microsoft account, with the multi-factor authentication and Conditional Access policies defined in Entra ID, and AWS permissions are assigned by group.

Identity is still a workstream in its own right, whichever cloud you pick: separate admin accounts, break-glass access, service accounts, joiner and leaver processes. Our founder worked on an identity overhaul on Azure for a major sports organisation, with Keycloak running on Kubernetes: projects like that show that identity architecture often matters more than the choice of provider.

3. Your team’s skills

A cloud nobody knows how to run is an expensive cloud. Take an honest inventory: do your administrators come from Windows and Active Directory, or from Linux and open source? Have your developers already shipped to one of the two platforms? Is your managed service provider comfortable with both?

Upskilling works in either direction, but it takes time and money. If you choose the platform your team knows least, budget for training and support from day one. Defining infrastructure as code (Terraform or OpenTofu, both usable on AWS and Azure) also reduces reliance on console clicks and makes knowledge transfer easier.

4. The managed services you will actually need

Start from your target architecture, not from the catalogue. List the essential components (managed PostgreSQL, Kubernetes, message queue, object storage, CDN, directory, backup) and check for each one:

• whether it is available in your target Region;
• its maturity (generally available or preview);
• the skills needed to run it;
• how it is billed.

For most standard needs, both platforms have an answer. Gaps appear on specific services: a particular database engine, an analytics tool, an integration with a line-of-business application. Those gaps, identified on your own list, are what should weigh in the decision.

5. Generative AI: Amazon Bedrock or Microsoft Foundry

Not long ago the dividing line was simple: OpenAI models on Azure, Anthropic models on AWS. That is no longer true.

• Amazon Bedrock is a fully managed service that, according to AWS, gives access to more than 100 foundation models, including from Amazon (Nova), Anthropic (Claude), DeepSeek, Moonshot AI, MiniMax, OpenAI and xAI.
• Microsoft Foundry (formerly Azure AI Foundry) brings agents, models and tools together under a single Azure resource. Microsoft cites more than 10,000 models from Microsoft, OpenAI, Anthropic, Meta and others. Existing Azure OpenAI resources can be upgraded to Foundry resources.
• On Google’s side, Vertex AI is now part of Gemini Enterprise Agent Platform, which offers Gemini models alongside partner models such as Claude and Mistral.

The choice of model no longer separates the platforms. Look instead at:

• Where processing happens. On Bedrock, cross-Region inference can be restricted to a geography (such as the EU) or set to global; AWS says global routing costs roughly 10% less but may process requests in any commercial Region. On Foundry, EU “Data Zone” deployments process data within Microsoft’s EU Data Boundary, while “Global” deployments may process it in any Azure region. For sensitive data, this setting matters more than the model name.
• Actual availability by Region. A model may be available globally before it reaches an EU zone. Check the exact availability before you design the architecture.
• Proximity to your data. If your documents sit in SharePoint and your users in Microsoft 365, the Microsoft ecosystem simplifies some use cases. If your application data is already on AWS, Bedrock avoids unnecessary transfers.

We cover running these services in production in our LLMOps approach.

6. Regions in France and compliance

All three providers have regions in France:

• AWS: Europe (Paris), eu-west-3, with three Availability Zones, enabled by default.
• Azure: France Central (Paris), with three availability zones, and France South (Marseille), its paired region, where access is restricted to specific scenarios such as disaster recovery.
• Google Cloud: europe-west9 (Paris), with three zones.

In France, personal health data must be hosted by an HDS-certified provider. AWS states it achieved HDS version 2 certification on 21 April 2026, with data at rest (backups included) required to reside in a European Economic Area Region, Paris among them. Microsoft states that an HDS 2.0 certificate was issued in October 2025, applying to Azure services listed as ISO/IEC 27001 compliant and deployed in regions including France Central and France South, with specific contractual terms to sign. In both cases the provider’s certificate does not cover your own configuration: encryption, permissions, logging and backups remain your responsibility. Our founder designs and runs AWS infrastructure in Terraform, with EKS, for health data hosting, so this is a subject we know from the inside.

7. Sovereignty

The word covers very different requirements. It helps to separate three levels:

• Data residency: data stays in a chosen Region or geography. All three providers support this. Microsoft completed its EU Data Boundary in February 2025, covering Microsoft 365, Dynamics 365, Power Platform and most Azure services.
• European operation: on 15 January 2026, AWS opened the AWS European Sovereign Cloud, with a first Region in Brandenburg, Germany. AWS describes it as physically and logically separate from other Regions and operated exclusively by EU residents. AWS also notes that it is not yet in scope of its HDS certification.
• SecNumCloud qualification from ANSSI, the French national cybersecurity agency, which is required for some sensitive public sector data in France. It relies on offerings run by French companies: S3NS, a Thales company working in partnership with Google Cloud, announced the qualification of its PREMI3NS offering in December 2025; Bleu, a joint venture between Capgemini and Orange built on Microsoft technology, was not yet qualified as of October 2026 and is targeting qualification by the end of the year.

If SecNumCloud is a hard requirement, the question is no longer “AWS or Azure” but “which qualified offering, and which services are actually available on it”. These statuses change quickly, so check them when you decide (information verified in October 2026).

8. Costs, discounts and commitments

Comparing list prices service by service is rarely conclusive: the gaps vary by service, Region and volume. What really drives the bill:

• Commitments. On AWS, Savings Plans commit you to an hourly spend for one or three years; AWS says you can save up to 72% on compute. On Azure, savings plans work on the same principle (hourly commitment over one or three years), and reservations offer deeper discounts on a specific resource when usage is stable.
• Licensing: Azure Hybrid Benefit on one side, the Listed Providers rules on the other (see criterion 1).
• Enterprise agreements: negotiated discounts, migration credits, partner programmes. They depend on your volume and your negotiation, and can only be compared case by case.
• Hidden costs: outbound data transfer, NAT gateways, logs kept for too long, non-production environments running around the clock.

The best indicator is not the provider that looks cheapest on paper, but the one whose bill you will be able to manage. That is what FinOps is about, as covered in our article on FinOps and AWS cost optimisation.

AWS vs Azure vs Google Cloud: comparison table

Criterion	AWS	Azure	Google Cloud
Typical strength	Breadth of catalogue, talent pool	Continuity with the Microsoft ecosystem	Data and analytics, Gemini models
Organisation	Accounts in AWS Organizations	Subscriptions under management groups	Projects in an organisation
Identity	IAM Identity Center, federated with Entra ID if needed	Native Entra ID	Cloud Identity, federation available
Managed Kubernetes	Amazon EKS	AKS	GKE
Generative AI	Amazon Bedrock	Microsoft Foundry	Gemini Enterprise Agent Platform
Region in France	Paris (3 zones)	France Central (3 zones), France South with restricted access	Paris (3 zones)
French health data (HDS)	HDS v2 certification (April 2026)	HDS 2.0 certificate (October 2025)	Scope to check with Google
Stronger sovereignty	AWS European Sovereign Cloud (Germany)	EU Data Boundary; Bleu offering under SecNumCloud qualification	S3NS PREMI3NS offering, SecNumCloud qualified
Pricing commitments	Savings Plans, Reserved Instances	Savings plans, reservations, Azure Hybrid Benefit	Committed use discounts

This table summarises trends and facts verified at the time of publication. It is no substitute for a review of your own list of services.

Multicloud: an option, not a goal

Multicloud is often presented as protection against lock-in. In practice, running two clouds properly doubles part of the work: two identity models, two networks, two sets of security tools, two bills to manage, and skills to maintain on both sides.

It makes sense in specific cases:

• a provider-specific service that a project cannot do without;
• a regulatory or contractual requirement;
• a merger or acquisition that brings two environments together;
• a deliberate, funded exit strategy.

On exit, two things have changed. First, all three providers now waive outbound internet data transfer fees for customers leaving their platform, subject to conditions (notifying support in advance, time limits, scope), in line with the direction set by the EU Data Act. Second, infrastructure defined as code (Terraform, OpenTofu) and containerised applications move far more easily than an environment built by hand. For most SMEs and mid-sized companies, one well-run primary cloud with a documented exit plan is a better trade-off than multicloud by accident. Our Kubernetes and Terraform work is built around exactly that portability.

Three typical situations

To make the framework concrete, here are three common profiles. They are illustrations, not rules.

• A mid-sized manufacturer, all-Microsoft. Identities in Entra ID, Microsoft 365, around thirty Windows servers and several SQL Server databases under Software Assurance. Azure starts with a head start thanks to licensing and tool continuity. AWS remains an option, but the licensing calculation comes first.
• A SaaS software vendor. Linux stack, containers, developers used to open source. Both platforms fit; the decision will rest on the team’s skills, the managed services needed and the customer base (some large customers mandate a provider).
• A healthcare organisation in France. Data subject to HDS, AI needed to work with documents. AWS and Azure are both certified; the decision will rest on where AI processing happens, the existing application estate and the team’s ability to secure the configuration.

How Feekr helps you decide

We work on both platforms and have no stake in steering you towards one or the other. Our approach is short and fact-based:

1. Scoping workshop: objectives, regulatory constraints, Microsoft estate, in-house skills, AI plans.
2. Inventory and estimate: the services you need, their availability in the target Regions, and a cost projection using the official calculators and your licences.
3. Decision note: a short document that justifies the choice, lists the risks and plans for exit.
4. Delivery if you want it: a landing zone defined as code, identity, networking, security, then migration.

Depending on the outcome, we support you on AWS or Azure, with cloud migration and cost control. If your decision is part of a wider migration programme, our article on cloud migration strategy and the 7 Rs complements this framework.

Next step

If you are torn between AWS and Azure, the most effective move is to test the framework against your real situation: your licences, your teams and your data. We can run that exercise with you from our base in Villeneuve-d’Ascq, on site in Lille and the surrounding area or remotely, and hand you a reasoned recommendation, whichever provider comes out on top.

[ YOUR QUESTIONS ]

Frequently asked questions

AWS or Azure: which should I choose?

There is no universal winner. Azure is often the natural choice when a company already runs on Microsoft (Entra ID, Microsoft 365, Windows Server, SQL Server) and wants to reuse its licences. AWS is frequently chosen for the breadth of its services, its maturity for cloud-native architectures and the size of its talent pool. The right answer depends on your estate, your teams, your compliance requirements and your AI plans.

What is the difference between AWS, Azure and Google Cloud?

All three provide compute, storage, managed databases, Kubernetes, serverless and AI services. The differences lie in the ecosystem (Microsoft for Azure), the history and breadth of the catalogue (AWS), strengths in data and analytics (Google Cloud, with BigQuery), and each provider’s positioning on generative AI. According to Synergy Research Group, AWS remains the largest cloud infrastructure provider worldwide, ahead of Microsoft and then Google.

Is Azure cheaper than AWS?

Neither is cheaper by default. Prices vary by service, Region and commitment (Savings Plans on AWS, savings plans and reservations on Azure). For a Windows Server or SQL Server estate covered by Software Assurance, Azure Hybrid Benefit can tip the balance towards Azure. The only reliable comparison is one built on your actual inventory, using both providers’ official calculators.

Can I use Microsoft Entra ID with AWS?

Yes. AWS IAM Identity Center supports SAML 2.0 federation and automatic provisioning of users and groups from Microsoft Entra ID through SCIM 2.0. Your staff sign in to AWS with their usual Microsoft account, and permissions are managed by group.

Should we go multicloud?

Not as a principle. Multicloud duplicates part of the skills, security tooling and governance. It makes sense when a specific need requires it: a provider-specific service, a regulatory requirement, an acquisition, or a deliberate and funded exit strategy. For most SMEs and mid-sized companies, one well-run primary cloud, defined as code, is the safer option.

Which cloud for French health data (HDS)?

AWS and Microsoft Azure are both HDS certified. AWS states it achieved version 2 of the framework in April 2026, with data at rest required to reside in European Economic Area Regions such as Paris. Microsoft states that an HDS 2.0 certificate was issued in October 2025, covering France Central and France South among others. Check the exact scope of the services you use and sign the required contractual terms.

Sources

• Synergy Research Group: Q2 Cloud Market Passes $143 Billion (30 July 2026)
• AWS: AWS Regions (Regions and Availability Zones)
• Microsoft Learn: List of Azure regions
• Google Cloud: Regions and zones (Compute Engine)
• AWS: Configure SAML and SCIM with Microsoft Entra ID and IAM Identity Center
• Microsoft Learn: Azure Hybrid Benefit (Azure SQL Database and SQL Managed Instance)
• Microsoft Licensing: New options for partner hosted cloud (Listed Providers)
• AWS: What is Amazon Bedrock?
• AWS: Cross-Region inference (Amazon Bedrock)
• Microsoft Learn: What is Microsoft Foundry?
• Microsoft Learn: Deployment types in Microsoft Foundry Models
• Google Cloud: Gemini Enterprise Agent Platform release notes (name changes from Vertex AI)
• AWS: HDS (Hébergeur de Données de Santé)
• Microsoft Learn: Health Data Hosting (HDS) France
• Microsoft: Microsoft completes landmark EU Data Boundary (26 February 2025)
• Amazon: AWS launches AWS European Sovereign Cloud (15 January 2026)
• Next: SecNumCloud, Bleu now targets qualification by year end (October 2026, in French)
• Business Wire: S3NS Announces SecNumCloud Qualification for PREMI3NS (December 2025)
• AWS: What are Savings Plans?
• Microsoft Learn: What are savings plans? (Azure)
• AWS News Blog: Free data transfer out to internet when moving out of AWS
• Microsoft Learn: Cancel your Azure subscription (free egress when leaving Azure)
• Google Cloud: Eliminating data transfer fees when migrating off Google Cloud