The Azure architect’s job is to turn competing requirements into a design that can actually be operated. A business might ask for an application that never goes down, retains customer data in a particular region, responds quickly worldwide and costs less than its existing environment. Those demands are not a shopping list of Azure services. They are constraints that interact: a second region improves some recovery options but adds cost and data-residency questions; private connectivity reduces exposure but complicates routing and troubleshooting. The Microsoft AZ-305 is concerned with this kind of decision-making.
AZ-305, Designing Microsoft Azure Infrastructure Solutions, is the current Azure architecture-design exam as of October 2026. It is not a newly announced replacement awaiting launch. The older AZ-303 and AZ-304 exams retired in March 2022. Microsoft’s English AZ-305 objectives were updated on April 17, 2026, and group the work into identity, governance and monitoring; data storage; business continuity; and infrastructure. The percentages in the current guide are useful for balancing preparation, but each area is connected to the others in real solutions. This article explains how to think through those decisions and how to prepare without treating older exam announcements as current advice.
Begin with constraints and failure conditions
A good architecture starts with conversations, not a reference diagram. Define the business transactions the system must complete, the data it must protect, the people who administer it, the latency its users can tolerate and the consequences of an outage. Ask whether the application is a new build, a migration or a hybrid service that must continue communicating with a data center. Clarify which requirements are mandatory and which can be negotiated when cost and complexity rise.
Express availability and recovery as measurable requirements. An acceptable recovery time objective tells the team how long a critical service may remain unavailable; a recovery point objective describes how much recent data loss the business can tolerate. These are not interchangeable. A system that recovers quickly from last night’s backup may violate the data-loss requirement, while a continuously replicated database can still take too long to restore application-level service if DNS, identity and dependent systems are neglected.
Record decisions and rejected alternatives. Why was a managed database chosen over a self-managed virtual machine? Why is an application active-passive instead of active-active? Which failure modes were tested? A decision log makes the eventual design auditable and prevents teams from assuming that a fashionable service automatically satisfies a contractual requirement.
Identity, governance and monitoring are design foundations
Identity is not an implementation detail to be added after networking. Applications need an answer to three different questions: who or what is authenticating, what actions are authorized and how access is reviewed. Microsoft Entra ID, managed identities, role-based access control and privileged access workflows solve different parts of that problem. For service-to-service access, a managed identity can remove a reusable credential from application configuration. For administrators, least-privilege role assignments, separate privileged accounts and reviewable elevation reduce the blast radius of a compromised identity.
Governance is equally architectural. Subscription structure, management groups, resource organization, Azure Policy, tagging and cost controls shape what teams can create and how an organization proves that its guardrails are working. A regulated team may need a policy preventing deployment outside permitted regions, while another team needs an exception process that is explicit and time-bound. A prohibition without an exception process often drives workarounds; an unrestricted exception undermines the policy.
Monitoring should be designed before production traffic arrives. Decide which application and platform signals indicate a genuine customer-impacting failure, where logs and metrics are stored, how long evidence is retained and which alert triggers an action. Diagnostic settings alone are insufficient if nobody owns the alert or knows which service boundary it represents. The associated Azure identity and governance design decisions must align with the operating team’s ability to investigate and respond.
Choose data services by workload behavior
Data design involves consistency, query patterns, throughput, schema changes, retention, access controls and geography. A transactional workload that needs relational joins and well-understood isolation differs from a document store optimized for flexible records or an object store for large unstructured files. A data platform that performs well in a benchmark can still be wrong if its operational model, recovery guarantees or compliance behavior contradict the application requirement.
Distinguish operational data from analytics. A customer-facing transaction database may need strict transactional consistency and predictable latency; analytics jobs can often accept asynchronous ingestion and work on copied or aggregated data. Mixing both workloads without resource and isolation planning creates avoidable contention. Where encryption, key management or private endpoints are required, confirm support at the chosen service tier instead of assuming feature parity across all offerings.
A storage design should explain why it uses replication and what that replication protects against. Regional replication addresses certain infrastructure failures but does not by itself prevent accidental deletion, malicious changes or application-level corruption. A separate backup and restore strategy may still be necessary. The useful question is not simply which storage account redundancy setting sounds safest, but which failures the organization needs to withstand and how the recovery will be demonstrated. Those trade-offs are examined in more detail in Azure data storage decisions.
Design recovery around the complete application
Business continuity is a system property, not a single replication checkbox. A web tier may run across availability zones while its database, authentication provider or message broker remains a single point of failure. A failover plan should map every dependency needed to complete an important user journey. Teams also need to distinguish a brief fault handled by retries from an extended zone or regional outage that requires an explicit recovery action.
Backups, replication, high availability and disaster recovery have different purposes. Backups preserve restorable historical states; replication can reduce data loss between healthy copies; availability zones separate some local failure domains; regional disaster recovery addresses broader outages. Designs can combine these, but they should not be used as synonyms. Retest whenever deployment architecture or critical data flows change. A recovery plan that worked before a new external integration was introduced may no longer meet the same objectives.
Failover should account for traffic steering, session state, capacity, DNS behavior, secret access and operational permissions. A passive region that has never carried a realistic workload may fail under load when needed most. The practical Azure business continuity design problem is to identify which assumptions can be rehearsed without causing customer harm and what evidence counts as successful recovery.
Make network and compute choices together
Infrastructure design spans compute models, network topology, connectivity, security boundaries, scalability and migration. Virtual machines provide operating-system control but place patching, image maintenance and more operational tasks on the team. Managed application platforms, containers and serverless components reduce some maintenance responsibilities while imposing their own runtime, networking and scaling constraints. Choose based on workload shape, operational capacity and the need for control, not a universal ranking of services.
Network decisions deserve the same rigor. Address ranges must avoid collisions across virtual networks, connected offices and future acquisitions. Hub-and-spoke structures, peering, routing, gateways and private endpoints affect traffic visibility and failure behavior. A workload that uses private access for a database still needs an intentional plan for outbound dependencies, name resolution and deployment tooling. A network security group can enforce traffic restrictions, but it is not a substitute for identity controls or application authorization.
When traffic spans regions or continents, assess where requests enter the system, which layers terminate TLS, which components can be cached and how the application handles failed downstream calls. Scaling a web service without considering connection limits, database saturation or message-processing capacity rarely fixes the real bottleneck. Azure infrastructure architecture involves these cross-layer relationships rather than isolated service descriptions.
Translate security incidents into architecture lessons
Historical stories about a large distributed denial-of-service attack can illustrate why architectural resilience matters, but an old incident should not be presented as part of the current AZ-305 announcement. DDoS protection, rate controls, traffic distribution, observability and application-level defenses address different risks. The correct answer depends on whether the threat is saturating network capacity, exhausting an application endpoint or exploiting an expensive downstream operation.
A resilient design considers both prevention and controlled degradation. An application might preserve account access and order history while temporarily limiting an optional report-generation feature. It might use caching, queues or protective thresholds to avoid turning an abnormal traffic spike into database exhaustion. Alerts should distinguish infrastructure saturation from ordinary seasonal demand, and runbooks should specify who can activate an emergency mitigation.
Security posture is a shared responsibility across the application and platform. Cloud services supply useful controls, but architects must still plan identities, encryption, data classification, audit trails, incident response and trusted deployment paths. A service advertised as secure does not absolve an application team of testing authorization checks or deciding what is logged.
Balance modernization, cost and maintainability
Cost analysis belongs in design reviews from the beginning. A cheaper compute instance may increase operational labor; an aggressive autoscaling plan may introduce connection pressure or unpredictable billing; a multi-region deployment may be required for resilience but unnecessary for a noncritical internal tool. Compare total operating cost, including data transfer, monitoring, storage growth, support, maintenance and the business consequence of downtime.
Migration plans should separate dependencies that can be rehosted safely from components that need redesign. A legacy application may require low-latency access to an on-premises service; moving its compute tier alone could make user transactions slower and less reliable. A phased migration, with measured telemetry and explicit rollback criteria, often creates better evidence than an all-at-once architectural promise. A clear migration and modernization plan ties each change to a testable outcome.
Maintainability matters long after an exam is over. Prefer repeatable deployments, coherent resource ownership, clear service boundaries and documented runbooks. Evaluate whether the people who must operate the solution can actually diagnose its failures and renew its security controls. A design that depends on one specialist’s undocumented workaround is fragile even when its architecture diagram looks elegant.
Prepare for AZ-305 as a design review
Microsoft’s April 17, 2026 English skills outline allocates roughly 25–30% to identity, governance and monitoring; 20–25% to data storage; 15–20% to business continuity; and 30–35% to infrastructure. Treat those weights as a map, not a substitute for the live official objectives. For each area, practice explaining why a proposed solution fits a requirement and why an attractive alternative fails one constraint.
Use scenarios that combine domains. An online retailer could require private database access, customer sign-in, regional recovery, structured audit evidence and a strict latency target. Design the solution once for its ordinary state, then test what happens when a region disappears, a key expires, a database reaches connection limits or an administrator’s access is revoked. Write down the signals that would reveal the failure and how the operators would respond. This exposes gaps that service-definition flashcards miss.
Azure Solutions Architect Expert also has a prerequisite certification requirement: verify the current Azure Administrator Associate prerequisite and the AZ-305 exam requirement on Microsoft Learn when planning the credential. If you arrive here through AZ-303 or AZ-304 materials, use them only as historical context; they do not replace current objectives or booking guidance. The key ability AZ-305 rewards is making a defensible design decision under constraints and explaining its operational consequences.