How do you know whether a cloud architecture is well designed? Since 2015, the industry answer has relied on AWS's Well-Architected Framework, which has been very widely adopted; Azure and Google Cloud have published close equivalents, but not identical ones. The AWS framework rests on 6 complementary pillars: operational excellence, security, reliability, performance efficiency, cost optimisation, and sustainability (added in 2021). This page details each pillar with its key questions and anti-patterns, presents the AWS / Azure / GCP frameworks, and introduces the recent EU Cloud Sovereignty Framework of October 2025, which covers the sovereignty dimension. It ends with a review template to hand in, which the final project reuses.
What you will learn
- The origin and evolution of the AWS Well-Architected Framework since 2015
- The 6 pillars in detail with their key questions and anti-patterns
- The equivalents: Azure Well-Architected and Google Cloud Well-Architected Framework
- The architecture resources of European sovereign providers
- The new EU Cloud Sovereignty Framework of October 2025
- A review template to hand in: a grid scored by pillar, evidence, decisions and an action plan
Prerequisites: know the IaaS, PaaS, SaaS, FaaS, CaaS grid and the notion of region and availability zone (in French). No operations experience is required: this lesson teaches you to read and score an architecture with a grid, not to build it. The final project of the course then has you apply this template to your own work.
1. Origin and evolution of the Well-Architected Framework
AWS published the white paper of the Well-Architected Framework in 2015 to formalise good architecture practices. The framework then had four pillars: security, reliability, performance efficiency, and cost optimisation. Operational excellence became a pillar in 2016, bringing the total to five. The sixth pillar, Sustainability, was added in December 2021 to measure and reduce the environmental impact of workloads.
The framework has been widely adopted by the industry, for three reasons: it is free and public (downloadable white papers), it offers an integrated tool (the Well-Architected Tool) that lets you assess an architecture, and it has established itself as a common vocabulary among architects, even outside AWS.
Microsoft published its Azure Well-Architected Framework in 2020 and keeps it at five pillars, with no sustainability pillar: Microsoft does publish Sustainability documentation, but presents it as cross-cutting guidance articulated with the five pillars, not as a sixth. Google Cloud publishes its Well-Architected Framework (originally called the Architecture Framework) and has six pillars, Sustainability included, with names of its own. IBM also publishes its own IBM Well-Architected Framework, centred on hybrid cloud and organised in six pillars. The convergence is real in substance, but it is not an identity: one more reason to check the framework of the provider you are auditing rather than assume.
2. The 6 Well-Architected pillars in detail
Pillar 1, Operational Excellence
Operational excellence is the ability to run and monitor systems to deliver business value, and to continuously improve processes and procedures.
Two questions help assess this pillar:
- How do you determine what the priority is when an incident occurs at 2 a.m., and is it documented in a runbook that the on-call engineer can follow without calling the tech lead?
- How often are your deployments automated, tested and rollback-able without manual intervention?
The typical anti-pattern combines manual deployments in production, runbooks in the head of the on-call person and no post-mortem after incidents: the organisation repeats the same mistakes for lack of formalised learning.
Pillar 2, Security
Security is the ability to protect data, systems and assets by relying on cloud technologies to improve the security posture.
Two questions help assess this pillar:
- How do you identify the users who access your resources, and are their permissions minimal in the sense of least privilege?
- How do you detect security incidents and how long does it take for a compromise to become visible in your dashboards?
The typical anti-pattern combines a root account shared between developers, security groups open to 0.0.0.0/0, no encryption at rest on sensitive data and CloudTrail logs disabled. See Shared responsibility for the detail.
Pillar 3, Reliability
Reliability is the ability of a system to perform its intended function correctly and consistently, despite expected failures.
Two questions help assess this pillar:
- What is the consequence of an AZ outage on your service? Do users see anything, or is it transparent?
- How often do you test your DR plan at full scale, and how long did you measure for the failover?
The typical anti-pattern combines a single-AZ architecture "because it is simpler", backups that are never tested by restoring and a DR runbook that asks for a password the on-call engineer does not have. See Cloud resilience: HA, DR, chaos engineering (in French).
Pillar 4, Performance Efficiency
Performance efficiency is the ability to use computing resources efficiently to meet the system requirements and to maintain that efficiency as demand changes.
Two questions help assess this pillar:
- How do you measure the performance perceived by your users, and can you tell p50 from p99 (the response time under which 50% then 99% of requests fall)?
- Are your resources sized correctly, or do you have instances running at 5% CPU 24/7 out of a habit inherited from on-prem?
The typical anti-pattern is monitoring that only measures CPU and RAM on the infrastructure side, never application latency on the user side. Added to that are systematic over-sizing "just in case" and no real benchmark before a scale-up.
Pillar 5, Cost Optimization
Cost optimisation is the ability to run systems that deliver business value at the lowest cost.
Two questions help assess this pillar:
- Do you know the breakdown of your bill by project, by team, by environment (dev / staging / prod)?
- What is the ratio of your consumption between on-demand, Reserved/Savings Plans, and Spot?
The typical anti-pattern is to stay 100% on-demand by default: commitment discounts, such as Savings Plans, Reserved Instances or Spot, are then left aside. Added to that are dev resources never stopped at night, a NAT Gateway provisioned per AZ without auditing the need and cumulative snapshots without a retention policy. See Pay-as-you-go: pricing models.
Pillar 6, Sustainability
Sustainability is the ability to minimise the environmental impact of cloud workloads, by optimising resource use and favouring low-carbon regions.
Two questions help assess this pillar:
- Do you know the energy mix of the region where your workloads run? Carbon intensity varies by an order of magnitude between a region with a mostly nuclear mix (France) and a region with a mostly coal mix (Poland); get the up-to-date value from your provider or from the energy agency.
- Are your non-urgent workloads (batch, ML training) scheduled in low-carbon regions when possible?
The typical anti-pattern combines deploying by default in AWS us-east-1 without any energy consideration, instances running 24/7 when usage is 8 am to 8 pm and cold data kept in Standard instead of Glacier. See Green cloud and sustainability (in French).
3. AWS vs Azure vs GCP comparison
The three frameworks cover essentially the same ground with slightly different vocabulary.
| Pillar | AWS Well-Architected | Azure Well-Architected | Google Cloud Well-Architected Framework |
|---|---|---|---|
| 1. Operational excellence | Operational Excellence | Operational Excellence | Operational Excellence |
| 2. Security | Security | Security | Security, privacy, and compliance |
| 3. Reliability | Reliability | Reliability | Reliability |
| 4. Performance | Performance Efficiency | Performance Efficiency | Performance optimization |
| 5. Cost | Cost Optimization | Cost Optimization | Cost optimization |
| 6. Sustainability | Sustainability | absent from the pillars | Sustainability |
As surveyed on 14 November 2026 in the official documentation, AWS and Google Cloud each have six pillars, Sustainability included, and Microsoft has five. Sustainability exists at Microsoft as cross-cutting recommendations, not as a pillar, which concretely changes what an Azure assessment tool will have you fill in.
The first five pillars overlap, the sixth does not, and two further differences deserve attention: Google groups security, privacy and compliance in a single pillar, whereas AWS and Azure isolate security, and it says Performance optimization where the other two say Performance Efficiency. The vocabulary is therefore largely shared, which lets you reason from one provider to another, but it is not interchangeable: an audit is conducted with the framework of the provider concerned, not with a universal grid. All three offer an interactive tool (AWS Well-Architected Tool, Azure Well-Architected Review, Google Cloud's review tool) that guides an assessment pillar by pillar.
4. Architecture resources at European sovereign providers
European sovereign cloud providers offer architecture resources in varied formats: technical guides, consulting services, resource centres. Here is a factual overview of the public resources in 2026, useful for an architect starting out with these providers.
Outscale publishes Technical Guides in its official documentation, covering good practices for high availability, security, and SecNumCloud compliance, as well as a Best Practices section (architecture within a Net, resilience, reversibility). Outscale does not publish a framework named "Well-Architected" (absence observed on docs.outscale.com on 12 October 2026): the Well-Architected Outscale series (in French) of this course applies the grid to this provider. The Outscale Cloud Practice programme offers support from Solutions Architects for complex deployments.
OVHcloud has opened an Architecture Centre that gathers guides, icons, reference architectures and a toolbox for architects. Official communication announces a foundational framework covering sustainability, automation and scalability, under development.
Scaleway offers an Architecture Review service delivered by its Solutions Architects, as well as detailed public blog resources on cloud-native architecture patterns.
Cloud Temple, Cegedim Cloud and other specialised sovereign providers offer tailor-made support adapted to the requirements of their segments (healthcare, defence, public sector).
How to use these resources with an existing framework
The vocabulary of the 6 Well-Architected pillars is widely shared and transposable. To structure an architecture review on a sovereign cloud, you can borrow the AWS / Azure / GCP grid and apply it to the equivalent services of the chosen sovereign provider. The concepts (Operational Excellence, Security, Reliability, Performance, Cost, Sustainability) remain valid regardless of the provider, with sustainability being a full pillar only at AWS and Google Cloud.
The recent EU Cloud Sovereignty Framework (see next section) brings a complementary dimension that the hyperscalers' WAFs did not have: legal and operational sovereignty.
5. EU Cloud Sovereignty Framework, the new piece of October 2025
In October 2025, the European Commission presented the Cloud Sovereignty Framework, a reference framework that assesses the sovereignty of a cloud solution. It is not a Well-Architected Framework in the AWS sense, it complements rather than replaces.
Structure of the framework
The framework defines 8 cloud sovereignty objectives, each graded on 5 levels (from 0 to 4) called SEAL (Sovereignty Effectiveness Assurance Level). An architecture can therefore be assessed objectively on each dimension, rather than being labelled in binary fashion "sovereign" or "not sovereign".
The Commission presented it for the Cloud III tender of the European Union institutions, then published an implementation guide titled "Cloud Sovereignty Framework, Implementation guidance". For compliance, objective SOV-7 (security and compliance) assesses adherence to GDPR, NIS2, DORA and other European frameworks; the CSRD (the European directive on corporate sustainability reporting, see section 7) is not listed there.
Why it complements the WAF
The AWS / Azure / GCP Well-Architected Frameworks deal with technical architecture excellence: performance, reliability, costs, operational security. The European Cloud Sovereignty Framework deals with legal and operational sovereignty: provider jurisdiction, effective control of data, technological independence, immunity to extraterritorial laws.
The two dimensions are distinct but complementary. An architecture can be excellent on the WAF side but weak on the Sovereignty Framework side (for example AWS on regulated European data). An architecture can be excellent on the Sovereignty side but immature on the WAF side (for example a sovereign deployment with no tracking of its service level indicators and objectives, SLI and SLO). The mature practice is to do both reviews on sensitive workloads.
For the detail of the sovereignty topic, see Technological sovereignty of the stack (in French).
6. How do you use these frameworks day to day?
The frameworks are not books to read once, they are operational tools used at 4 distinct moments in the life of an architecture.
At initial design
Before the first line of Terraform, organise a design session where you walk through the 6 WAF pillars with the team. For each pillar, identify 2 to 3 structuring decisions. Document the choices in an Architecture Decision Record (ADR) versioned in Git. Plan half a day for a medium-sized architecture (the author's benchmark, to adjust to your team): this time often avoids a long rework a few months later.
If the architecture targets a sovereign cloud and handles regulated data, add a SEAL session on the 8 objectives of the Cloud Sovereignty Framework on the same day.
At a pre-production review
Before going into production, organise a Well-Architected review with an architect not involved in the initial design. The AWS Well-Architected Tool is a review questionnaire: the team answers the questions of each pillar and the tool derives a list of risks and an improvement plan from them. It does not analyse the deployed infrastructure. Azure Advisor and Google Cloud Active Assist, on the other hand, analyse the existing resources and issue automatic recommendations. In both cases, the result is a starting point: a human must then prioritise according to the real business risk.
On a sovereign cloud with no integrated tool, run the review with an in-house questionnaire template or one borrowed from public frameworks. It is less industrial but perfectly viable.
During an incident
When an incident occurs, the framework helps classify the root cause on the relevant pillar. "This outage is due to Reliability (single-AZ)" is more actionable than "we had an outage". The incident becomes a learning opportunity that strengthens the identified pillar. Combined with blameless post-mortems (incident reviews that look for the cause, not for someone to blame), it is a weapon against repeated mistakes.
During an annual audit
At least once a year, run a full Well-Architected review across all critical architectures. It is the occasion to measure the change since the last review, to prioritise the projects of the coming year, and to calibrate the cloud budget according to the needs identified. Without this annual discipline, the framework becomes a dusty binder that nobody opens.
7. How do you use the framework depending on context?
The framework does not impose a uniform weighting of the 6 pillars; each organisation calibrates its priorities according to its business, its constraints and its life cycle. Here is how the pillars are typically prioritised depending on the context.
In the start-up and early phase, the Cost Optimization and Performance Efficiency pillars come first. Sustainability and Reliability are at a baseline level, to be strengthened once the product has found its traction. Do not over-invest in multi-region HA until the market is validated.
In growth and scale-up, the priority shifts to Reliability and Operational Excellence. Outages now have a measurable business cost, and the organisation must industrialise its operational practices.
In a regulated sector (banking, healthcare, government), Security and Reliability dominate in general. Regulatory constraints and contractual SLA requirements call for architectures across several availability zones from the start, a recovery plan in another region sized on the required recovery objective, and audit trails kept from day one.
At maturity, when the organisation takes on CSR commitments and comes within the scope of the European CSRD directive (the sustainability reporting of companies), Sustainability gains importance. Since Directive (EU) 2026/470, known as Omnibus I, the CSRD only targets companies with more than 1,000 employees and more than EUR 450 million in net turnover. Its timetable depends on transposition in each Member State (status on 13 November 2026). See Green cloud and sustainability (in French) for the detail.
8. Which template should you follow to run an architecture review?
An architecture review is run with a grid scored by pillar, where each score rests on a piece of evidence, and it ends with a written deliverable. Without these three elements, the review remains a conversation of which nobody keeps a trace. This template is a working proposal inspired by the providers' review tools, not a published standard: adapt it to your organisation.
Which inputs to gather before the review?
Gather four elements before the first question: the up-to-date architecture diagram, the ADRs already written (see section 6), the quantified requirements the architecture must meet (target availability, monthly budget, sovereignty constraints) and the list of components with their owner. A review without quantified requirements can only judge the team's comfort: you do not score an architecture as "reliable" without saying reliable at what level.
How do you score each pillar?
The grid has one axis per row: the first five pillars (operational excellence, security, reliability, performance, cost) and a sixth axis, sovereignty, scored with the objectives of section 5. Sustainability is added as a seventh row if your organisation tracks it. For each axis, ask one question (the final project gives one per axis, otherwise take the more exposed of the two questions of section 2), then score from 0 to 3, asking for a piece of evidence each time. A score of 0 means that no document covers the axis. As soon as a document exists, the score is at least 1: it is 1 if this document shows that the requirement is not met, or if a point of the requirement has not been assessed.
| Score | Meaning | Expected evidence |
|---|---|---|
| 0 | no document covers the axis | none |
| 1 | a piece of evidence exists and shows that the requirement is not met, or a point of the requirement has not been assessed | a finding, a measured gap, or the mention of the point not assessed |
| 2 | the evidence shows that the requirement is met, without verification | configuration, estimate, plan or written runbook |
| 3 | a verified and dated test or measurement shows that the requirement is met | result of a test, an exercise or a dated measurement |
Example on the Reliability pillar: "the application runs on two zones" scores 2 if the configuration shows it, 3 only if a zone outage was simulated and the service stayed available. A score of 3 therefore requires a measurement, not an intention.
What must the deliverable contain?
The deliverable fits in one to two pages: the table of scores (one per axis) with the evidence retained for each, the three most serious risks, ranked by severity multiplied by probability (each scored on a small scale, the product gives the rank), a decision per risk (fix, accept, postpone) recorded in an ADR, and an action plan with an owner and a deadline. The decisive point is the decision: an accepted risk must be signed by someone who bears the consequence, otherwise it is simply forgotten.
On a sovereign cloud, add as an appendix the detail of the SEAL levels obtained on the eight objectives of section 5, whose summary feeds the sovereignty row. The final project of the course has you hand in a review of this type on your own architecture.
Put it into practice
The lab provides an evidence pack on the architecture of a portal, with its risks and the limits of that evidence. You write two files: review.json, which scores the six axes from 0 to 3 citing one piece of evidence per axis, retains the three most serious risks with a decision, an owner and a deadline, and declares the limits of the evidence; then review.md, a short report for the architecture board. The validation applies the lab's grid to the evidence, according to its type and content. It checks that the three risks are those given by the severity times probability calculation, in order, that each decision is assigned, that all the limits are declared and that the report covers the six axes and the three risks. Nothing is assessed on a real cloud account.
Key points
- AWS's Well-Architected Framework (2015) has become the common reference of cloud architecture, with Azure (5 pillars) and Google Cloud (6 pillars) publishing close but not identical equivalents.
- 6 pillars at AWS and Google Cloud: Operational Excellence, Security, Reliability, Performance Efficiency (Performance optimization at Google), Cost Optimization, Sustainability. Azure has only 5, sustainability being cross-cutting guidance.
- Each pillar contains structuring questions and anti-patterns identified by the industry.
- European sovereign providers (Outscale, OVHcloud, Scaleway, Cloud Temple) have not all published a formal equivalent: Technical Guides, Architecture Centre, Architecture Review exist, the Well-Architected Outscale series (in French) of this course applies the AWS grid to Outscale, but there is no unified framework common to sovereign providers.
- The EU Cloud Sovereignty Framework of October 2025 complements the WAFs by covering legal and operational sovereignty (8 objectives, SEAL levels 0 to 4).
- The frameworks are used at 4 moments: initial design, pre-prod review, incident post-mortem, annual audit.
- A review is handed in as a grid scored by axis (five pillars plus sovereignty, 0 to 3, one piece of evidence per score, 0 when no document covers the axis), three risks ranked by severity times probability, decisions recorded in ADRs and an action plan.
- Shared vocabulary is the main contribution of the frameworks, often more useful than the technical checklist.
Knowledge check
Check that the essentials of this guide are in place. The questions only cover what has just been explained here.
Knowledge check
Check what you have learned with this interactive quiz
How it works
- The timer starts when you click Start
- Multiple choice, true/false and short answer questions
- You can move between questions
- Detailed results are shown at the end
Starts the quiz and the timer
Review
(0/0)Score by domain
A domain groups several guides and gets several questions, so its score is more reliable than a per-topic score, which often rests on a single question.
Summary
Strengths
No topic passed at 80% on at least two questions yet.
Weak areas
No topic below the pass mark.
Recommended next
Hints before you try again?
Continue the course: Design a target architecture: compare and decide
A new full quiz with random questions
Work again only on the questions you missed