In this article
Cloud basics
What is cloud computing?
Cloud computing is a model for getting on-demand network access to a shared pool of configurable computing resources that can be provisioned and released with little management effort. The National Institute of Standards and Technology (NIST) Special Publication (SP) 800-145 lists five essential characteristics: on-demand self-service, broad network access, resource pooling, rapid elasticity and measured service (NIST SP 800-145, September 2011).
Why it is asked: it opens the DevOps and cloud topics and checks you can define the term precisely, not by naming a vendor.
What is the difference between IaaS, PaaS and SaaS?
Infrastructure as a Service (IaaS) rents basic computing, storage and networking; Platform as a Service (PaaS) lets you deploy your own application onto a provider's platform; Software as a Service (SaaS) is a finished application you use. NIST defines all three, and the consumer controls less of the stack with each step from IaaS to SaaS (NIST SP 800-145, September 2011).
Why it is asked: it frames every later question about who operates what.
What is the shared responsibility model?
The shared responsibility model is the split of security duties between the cloud provider and the customer. Amazon Web Services (AWS) describes its part as security of the cloud, the infrastructure that runs its services, and the customer's part as security in the cloud, which varies by service: on a virtual machine the customer manages the guest operating system and firewall rules, while for managed storage the customer manages data, encryption options and permissions (AWS shared responsibility page, read October 2026).
Why it is asked: it shows you know which security tasks stay with the customer.
Design principles
What is a Well-Architected Framework?
A Well-Architected Framework is a provider's set of design pillars and questions for judging whether a workload is built well. AWS names six pillars: operational excellence, security, reliability, performance efficiency, cost optimization and sustainability (AWS Well-Architected pillars, read October 2026). Azure lists five and Google Cloud six, so quote the vendor you interview with.
Why it is asked: it is the shared vocabulary of architecture reviews.
How do the Azure and Google frameworks differ from AWS's?
They overlap heavily. Azure's framework has five pillars (reliability, security, cost optimization, operational excellence and performance efficiency), with no sustainability pillar, and pairs each with trade-off guidance (Microsoft Learn, updated January 2025). Google Cloud's six map one to one onto AWS's, but fold privacy and compliance into security and call the performance pillar performance optimization (Google Cloud, reviewed October 2026).
Why it is asked: it shows you can work across providers.
How do you handle trade-offs between pillars?
Rank the pillars for the workload, then decide per component. A payments service puts reliability and security first and accepts higher cost; an internal batch job accepts lower reliability to cut cost. Azure's framework gives each pillar a trade-offs page for exactly this reason (Microsoft Learn, updated January 2025). Write the choice and its reason down.
Why it is asked: architects are paid for judgment, not for listing pillars.
Why design components to be stateless and decoupled?
A stateless component keeps no user data between requests, so any copy can serve any request and copies can be added or lost without harm. Google Cloud's core principles include using a stateless architecture, relying on shared storage or cache rather than local state, and decoupling the system into components that can be upgraded and monitored independently (Google Cloud, reviewed October 2026).
Why it is asked: it underlies scaling and recovery answers.
Knowing it is half the job. Saying it is the other half.
Interview Clozo is an AI interview copilot for real interview rounds: it hears the question and puts a strong answer on your screen, and you say it in your own words. See the features or how it works.
Start Free Trial →Reliability and disaster recovery
What are RTO and RPO?
The recovery time objective (RTO) is the longest downtime the business accepts; the recovery point objective (RPO) is how much recent data may be lost, measured as time since the last good copy. AWS ties the backup schedule to the achievable recovery point and the speed of redeployment to the recovery time (AWS whitepaper, read October 2026).
Why it is asked: every disaster recovery design starts from these two numbers.
What are the four disaster recovery strategies?
AWS describes four, from cheap to costly: backup and restore, pilot light, warm standby and multi-site active/active. A pilot light is a replicated copy of the data plus core infrastructure, with application servers switched off; warm standby is a scaled-down but fully working copy that can serve traffic straight away at reduced capacity; active/active serves traffic from several regions (AWS whitepaper, read October 2026).
Text description of the diagram
- Pilot light keeps data replicated and core infrastructure ready but application servers switched off.
- Warm standby keeps a scaled-down working copy running.
- Active/active is the most complex and costly approach.
Why it is asked: it tests whether you can match cost to the recovery target.
How do you choose between multi-zone and multi-region?
Choose by what failure you must survive. A region is a separate geographic area; a zone is an isolated group of data centers inside a region. AWS notes that for a well-architected, highly available workload, losing one data center may need only backup and restore, while a regional disaster, or a regulation that demands it, points to pilot light, warm standby or active/active (AWS whitepaper, read October 2026). Multi-region also brings data-consistency and cost problems.
Why it is asked: it separates over-engineering from a defensible design.
Why is active/active not always the answer?
Active/active is the most complex and costly disaster recovery approach, even though it can cut recovery time to near zero for most disasters. AWS adds that data corruption may still need backups, so the recovery point is not zero, and that you must design how writes stay consistent across regions (AWS whitepaper, read October 2026).
Why it is asked: restraint with expensive designs is a senior signal.
Why should failover avoid control-plane operations?
A control plane is the set of operations that configure the environment, while the data plane delivers the live service. AWS says data planes typically have higher availability design goals, so for maximum resilience a failover should use only data-plane operations (AWS whitepaper, read October 2026). A plan that must create resources during the outage depends on the less available part.
Why it is asked: it tests depth beyond standard diagrams.
Should failover be automatic?
Use caution. AWS warns that even with good practice recovery time and recovery point are greater than zero, so an unnecessary failover after a false alarm costs availability and data, and manual initiation is therefore common. It still recommends automating the steps so the manual trigger is like pushing a button (AWS whitepaper, read October 2026).
Why it is asked: it probes your view of operational risk.
How would you design a highly available web application?
Clarify the availability target, then place stateless application servers across several zones behind a load balancer, keep data in a replicated managed store, define the recovery time and recovery point, and automate deployment with infrastructure as code (defining infrastructure in version-controlled files). Say what failure each choice survives and what it costs. The same reasoning applies to machine learning system design, where model serving adds graphics processing unit (GPU) capacity to the plan.
Why it is asked: it combines the topics on this page in one open design prompt.
Cost, scaling and security
How do you design for scale: scale up or scale out?
Scaling up gives one machine more power; scaling out adds more machines behind a balancer. Azure's performance principle is to scale horizontally, test early and often, and monitor the health of the solution (Microsoft Learn, updated January 2025). Scaling out needs stateless components and shared state, which is why the two questions go together.
Why it is asked: it links scaling to application design.
How would you reduce a cloud bill?
Remove waste first, then reduce the rate you pay. Azure frames cost optimization as optimizing on usage and on rate utilization while keeping a cost-efficient mindset (Microsoft Learn, updated January 2025). In practice: switch off idle resources, right-size what remains, then use commitment discounts on steady load. Measure with budgets and tags before changing anything.
Why it is asked: cost is a first-class architecture concern.
How do you secure a workload in the cloud?
Start from the shared responsibility model: know which layers the provider secures and which you must. AWS says the customer manages data, encryption options and permissions for managed services, and the guest operating system and firewall rules for virtual machines (AWS shared responsibility page, read October 2026). Add least-privilege access, encryption, logging and regular review.
Why it is asked: security answers show whether you can name the layers you own.
Why use infrastructure as code?
Infrastructure as code (IaC) is defining infrastructure in version-controlled files that tools deploy repeatably instead of clicking in a console. AWS says deploying with IaC lets infrastructure be redeployed quickly and without errors in a recovery region, and that without it restoring a workload becomes complex and may exceed the recovery time objective (AWS whitepaper, read October 2026).
Why it is asked: repeatability is the base of reliable operations.
Migration
What are the seven Rs of cloud migration?
The seven migration strategies are retire, retain, rehost, relocate, repurchase, replatform and refactor or re-architect. AWS Prescriptive Guidance defines each: rehost, also called lift and shift, moves an application unchanged; replatform adds some optimization; refactor redesigns it for the cloud (AWS Prescriptive Guidance, read October 2026).
Why it is asked: it is the standard vocabulary of migration plans.
When would you refactor instead of rehost?
Refactor when the application's design is itself the problem, such as a monolith slowing delivery or a legacy system nobody can maintain. AWS calls refactoring the most complex and costly strategy and recommends it for large migrations only when other options are unacceptable, modernizing after the move instead (AWS Prescriptive Guidance, read October 2026).
Why it is asked: it tests sequencing and risk judgment.
Follow-up questions
What are the most common cloud architect interview questions?
The core topics are cloud basics and the shared responsibility model, design trade-offs, disaster recovery, cost, security and migration, and a design round adds an open scenario prompt such as a highly available web application. The wider structure of rounds is in interview loops.
How do I prepare for a cloud architect interview?
Draw three architectures you can defend, such as a web application, a data pipeline and a disaster recovery plan, and rehearse the trade-offs aloud. Read one provider's framework end to end, then prepare your introduction and questions for the team (questions to ask the interviewer); behavioral interviews cover the people side.
Do I need a cloud certification to be a cloud architect?
Certifications show you studied a provider's services, but interviewers usually test whether you can reason about a design. Prepare for both: the vocabulary a certification teaches helps, and a worked design with trade-offs proves judgment.
- Define cloud terms from a standard such as NIST, not from one vendor.
- Match the disaster recovery strategy to the RTO and RPO the business needs.
- State the trade-off between pillars for each design decision.
- Avoid control-plane dependencies in a failover plan.
- Sequence a migration: move first, modernize after, unless the design is the problem.
Sources
- The NIST Definition of Cloud Computing (SP 800-145) — NIST (Mell and Grance), 28 Sep 2011
- The pillars of the AWS Well-Architected Framework — Amazon Web Services (date = read date), 24 Oct 2026
- Microsoft Azure Well-Architected Framework: pillars — Microsoft Learn, 22 Jan 2025
- Google Cloud Well-Architected Framework — Google Cloud (last reviewed), 13 Oct 2026
- Shared Responsibility Model — Amazon Web Services (date = read date), 24 Oct 2026
- Disaster recovery options in the cloud — Amazon Web Services whitepaper (date = read date), 24 Oct 2026
- About the migration strategies (the 7 Rs) — AWS Prescriptive Guidance (date = read date), 24 Oct 2026
