AWS Certified Solutions Architect – Professional (SAP-C03)
The hardest exam in the roadmap. It covers complex, multi-account, hybrid, secure, resilient and AI-integrated architectures.
Verified against official sources on 2026-11-17.
AWS Certified Solutions Architect – Professional (SAP-C03) checks that you can design optimised AWS solutions across many applications and accounts in a complex organisation. That covers architectures built for the cloud, security and governance, cost, resilience and migration, and operational excellence. The new version includes AI and ML integration. Passing it renews SAA-C03.
Brand-new version
SAP-C03 has been used from 2026-11-17 (today, at the time of writing). SAP-C02 ended on 2026-11-16. Most third-party courses still teach C02. Its domain structure is completely different, so build your study plan from the C03 exam guide.
At a glance
| Code | SAP-C03 |
| Level | Professional |
| Intended audience | 2 or more years designing and implementing AWS solutions, giving architectural guidance across multiple applications in a complex organisation |
| Questions | 75 (65 scored + 10 unscored) |
| Time | 180 minutes |
| Passing score | 720 (scaled 100–1000) |
| Price | USD 300 |
| Question types | Multiple choice. Multiple response (you must select every correct option) |
| Valid for | 3 years. Renew by passing the latest SAP exam, or with the paid Skill Builder maintain option |
| Out of scope | In-depth OS knowledge, ML model training and fine-tuning, detailed security forensics, writing and debugging application code |
Official domains and weightings
| Domain | Weight |
|---|---|
| 1. Architecture Design and Implementation Built for the Cloud | 25% |
| 2. Security, Compliance, and Governance | 25% |
| 3. Cost-Optimized Architecture Design | 15% |
| 4. Resilience, Migration, and Business Continuity | 15% |
| 5. Operational Excellence and Automation | 20% |
Main additions compared with SAP-C02: Amazon Bedrock, RAG and AI agent architectures (including AgentCore), SageMaker AI inference, durable multi-step workflows, VPC Lattice and ECS Service Connect, multi-tenant isolation, IPv6 and dual-stack, lakehouse architectures with Iceberg, AWS Clean Rooms, IAM Roles Anywhere, post-quantum cryptography in KMS (ML-DSA, ML-KEM), data residency controls, compliance evidence with AWS Artifact, threat modelling, Bedrock Guardrails, human approval for AI operations, cost optimisation for AI workloads, and DevSecOps pipeline scanning and signing. IoT, end-user computing and media services were removed from the in-scope list.
Key AWS services tested
- Multi-account and governance: Organizations, SCPs and RCPs, Control Tower, IAM Identity Center, RAM, Config, Security Hub, Security Lake, Audit Manager, AWS Artifact
- Networking: Transit Gateway, Cloud WAN, Direct Connect (gateways, VIF types), Site-to-Site VPN, PrivateLink, VPC Lattice, Route 53 Resolver, Network Firewall, CloudFront, Global Accelerator
- Compute and integration: ECS, EKS, Lambda, Step Functions, EventBridge, SQS, SNS, API Gateway (including WebSocket), Kinesis, MSK
- Data: Aurora (Global Database), DynamoDB (global tables), RDS, ElastiCache, MemoryDB, Redshift, S3, Lake Formation, Glue, Athena, EMR, Clean Rooms
- AI: Amazon Bedrock (Knowledge Bases, Guardrails, AgentCore, Evaluations; Agents Classic is closed to new customers since 2026-07-30), SageMaker AI endpoints
- Resilience and migration: AWS Backup, Elastic Disaster Recovery, Application Migration Service, DMS, AWS Transform, Fault Injection Service, Resilience Hub, Snow Family, DataSync
- Operations: CloudFormation, CDK, CodePipeline, Systems Manager, CloudWatch (Container Insights, Synthetics, RUM), X-Ray, the OpenTelemetry distro
What you already know vs what you must learn
| You already know | You must learn |
|---|---|
| Enterprise integration with core banking, MQ and file transfer | Hybrid network design at scale: Direct Connect plus Transit Gateway, centralised inspection, DNS across on-premises and AWS |
| Large-scale system design and trade-off thinking | Choosing the best of four plausible AWS designs under constraints (cost, RTO and RPO, operational effort) |
| .NET Framework migrations and SQL Server estates | Migration strategy (the 7 Rs), MGN, DMS, AWS Transform for .NET, Babelfish |
| Regulatory constraints (data residency, audit) | Governance at organisation scale: SCPs, RCPs, data perimeters, Region restrictions, evidence collection |
| A little GenAI from Module 16 | AI architectures: RAG with Bedrock Knowledge Bases, agents, guardrails, human approval |
Preparation order and study plan (about 80–120 hours over 8–12 weeks, last in the roadmap)
| Week | Focus | Hours |
|---|---|---|
| 1 | Exam guide and the C02→C03 comparison. Mark every new task | 6 |
| 2–3 | Domain 1: Module 14 banking architectures, Module 02 hybrid networking, Module 07 lakehouse, Module 16 GenAI | 20–25 |
| 4–5 | Domain 2: Module 05 and Module 15 at organisation scale. Data residency. Post-quantum KMS basics | 15–20 |
| 6 | Domain 3: Module 15 FinOps, cost of AI workloads, observability costs | 8–10 |
| 7 | Domain 4: Module 12 DR (multi-Region), Module 13 migration | 10–12 |
| 8 | Domain 5: Module 09, 10 and 11 at organisation scale (multi-account pipelines, observability) | 8–10 |
| 9–12 | Free Exam Prep course, the 20-question set, this page's questions, timed long-scenario practice, mock exams | 15–30 |
Mapped platform modules
- Module 14 — Banking architecture: Domain 1 scenarios
- Module 02 — Networking: hybrid and multi-VPC
- Module 05 — Security and Module 15 — Governance & FinOps: Domains 2 and 3
- Module 12 — HA & DR and Module 13 — Microsoft migration: Domain 4
- Module 09 — DevOps, Module 10 — IaC and Module 11 — Monitoring: Domain 5
- Module 07 — Data engineering and Module 16 — GenAI: data and AI architectures
Practical skills to build
- Design (on paper, or deploy in a small form) a hub-and-spoke network: Transit Gateway, inspection VPC with Network Firewall, centralised egress, Direct Connect to a simulated on-premises network
- Write a multi-Region DR design for a payments platform with RTO and RPO targets, and run a failover drill using Aurora Global Database switchover
- Build a Region-restriction SCP with exemptions for global services, and test it
- Build a small RAG assistant over fictional policy documents with Bedrock Knowledge Bases, metadata filters and a guardrail
- Write a migration plan for a .NET Framework 4.8 application with SQL Server: assess, rehost, then replatform and refactor
Exam-prep checklist
- I can choose between Transit Gateway, Cloud WAN, peering, PrivateLink and VPC Lattice for a scenario
- I know Direct Connect resiliency models and how to combine DX with VPN
- I can match DR strategies to RTO and RPO and cost, using Aurora Global Database, DynamoDB global tables, Elastic Disaster Recovery and AWS Backup
- I can design org-wide guardrails: SCPs, RCPs, data perimeters, Region restrictions, delegated administrator accounts
- I can design RAG and agent architectures with access control and guardrails
- I know the 7 Rs and when to use MGN, DMS, AWS Transform and Babelfish
- I can optimise cost for compute (Savings Plans, Graviton, Spot), storage, data transfer and AI inference
- I can read a long scenario, pick out the deciding constraint, and eliminate wrong options in about 2.4 minutes
Official links
- Exam guide (SAP-C03)
- Comparison of SAP-C02 and SAP-C03
- Exam product page
- Skill Builder Exam Prep plan (SAP-C03)
Practice questions
These 5 questions are original, banking-flavoured scenarios. They aren't real exam questions.
A bank has 40 AWS accounts, each with VPCs that need private connectivity to an on-premises core banking system over its existing Direct Connect link. Security requires all traffic between VPCs and on-premises to pass through a central inspection point. The network team wants to manage as few connections as possible. What design fits best?
- A. A full mesh of VPC peering connections, plus a private VIF from every VPC
- B. A Transit Gateway shared through AWS RAM, with Direct Connect connected through a Direct Connect gateway and a transit VIF, and an inspection VPC running AWS Network Firewall. Route tables send VPC and on-premises traffic through inspection, with appliance mode turned on for the inspection attachment
- C. A separate Site-to-Site VPN from every VPC to on-premises
- D. A PrivateLink endpoint service for every application
Correct answer: B
Explanation: Transit Gateway is a regional hub that routes between VPCs and to on-premises (transitive routing). Sharing it through RAM lets every account attach to it. A transit VIF on a Direct Connect gateway connects DX to the TGW. Separate TGW route tables force traffic through the inspection VPC. Appliance mode keeps both directions of a flow on the same firewall endpoint, so stateful inspection works.
Why the others are wrong:
- A: VPC peering isn't transitive. A mesh for 40 or more VPCs is hard to manage, and there's no central inspection point.
- C: Dozens of VPN tunnels means heavy operations, limited bandwidth, and no central inspection.
- D: PrivateLink exposes specific services. It doesn't provide general routed connectivity or inspection.
A payments database runs on Aurora PostgreSQL in one Region. The business sets RPO in seconds and RTO under 15 minutes if the Region fails. Cost matters, so a fully active-active design isn't wanted. What should the architect recommend?
- A. Copy automated snapshots to the second Region every hour
- B. Aurora Global Database, with a secondary cluster in the DR Region and warm-standby application capacity there (scaled down, ready to scale up). Use a tested managed failover runbook, and Route 53 for endpoint switching
- C. Aurora Multi-AZ in the primary Region only
- D. Multi-Region active-active with application-level conflict resolution between two writable PostgreSQL clusters
Correct answer: B
Explanation: Aurora Global Database replicates at the storage layer with typical lag under a second, which meets an RPO in seconds. Promoting the secondary takes minutes. Warm-standby compute (for example, ECS services at minimum capacity) meets the RTO without paying for full duplicate capacity. Test it: run game days, measure the RTO, and record the evidence for regulators who expect DR to be tested.
Why the others are wrong:
- A: Hourly snapshots mean an RPO of up to an hour, and restoring takes longer than 15 minutes.
- C: Multi-AZ protects against an AZ failure, not a Region failure.
- D: Active-active with multiple writers is expensive and complex, and conflict resolution is risky for money movements. It's more than the requirements need.
The bank's legal team says customer workloads must run only in eu-west-2 (London). The cloud team must prevent resources from being created in any other Region across all workload accounts, but global services such as IAM, Route 53 and CloudFront must keep working. What's the best control?
- A. An IAM policy on every role that denies actions outside
eu-west-2 - B. An SCP on the workload OUs that denies actions when
aws:RequestedRegionisn'teu-west-2, usingNotActionto exempt global services (or turn on the equivalent Control Tower Region deny control) - C. An AWS Config rule that flags resources created in other Regions
- D. Tag policies that require a
region=eu-west-2tag
Correct answer: B
Explanation: An SCP is a preventive control across all accounts in the OU, and administrators inside those accounts can't override it. The aws:RequestedRegion condition key restricts Regions. Global services have to be exempted, because their API calls go to us-east-1 endpoints. Control Tower offers a managed version of this control. Document it as a residency control, but legal and compliance still decide what residency actually requires (backups, support access, logs).
Why the others are wrong:
- A: IAM policies on each role can be changed by account administrators, and new roles can be created without them.
- C: Config only detects problems after the fact. It doesn't prevent them.
- D: Tags are metadata. A tag doesn't control where a resource is created.
The bank wants an internal assistant that answers staff questions from policy documents stored in S3. Requirements: answers must be grounded in the documents with citations, staff must see answers only from documents their department is allowed to read, and PII must not appear in responses. What architecture fits?
- A. Fine-tune a foundation model on all policy documents and expose it to all staff
- B. Amazon Bedrock Knowledge Bases (RAG) over the S3 documents, with department metadata on each document and metadata filtering at query time based on the caller's identity, plus a Bedrock guardrail with sensitive-information filters. Log invocations for audit
- C. Put every document in the system prompt on every request
- D. Train a custom large language model on SageMaker AI from scratch
Correct answer: B
Explanation: RAG retrieves the relevant passages and passes them to the model, so answers are grounded and can cite sources. When documents change, you re-sync the knowledge base without retraining. Metadata filtering, driven by the authenticated user's department claims, makes sure retrieval returns only allowed documents. Enforce this on the server, never by trusting what the client sends. Guardrails mask or block PII. The bank's model-risk, privacy and security teams still need to review the design.
Why the others are wrong:
- A: Fine-tuning mixes all documents into the model with no access control per department, gives no citations, and becomes stale.
- C: Large prompts hit context limits, cost a lot per request, and give every user every document.
- D: Training from scratch is very expensive and solves none of the stated requirements.
A bank must leave its data centre within 4 months. One critical application is a .NET Framework 4.8 web app on Windows Server with SQL Server. Leadership also wants to cut Windows and SQL Server licensing costs over the next 2 years. Which strategy balances speed and the long-term goal?
- A. Rewrite the application as .NET 10 microservices on Lambda and Aurora PostgreSQL before migrating
- B. Rehost first with AWS Application Migration Service (MGN), and move the database to Amazon RDS for SQL Server (with DMS if needed) to meet the deadline. Then replatform or refactor in stages: port to modern .NET on Linux containers (for example, with AWS Transform for .NET), and evaluate Aurora PostgreSQL with Babelfish to reduce SQL Server licensing
- C. Retain the application on-premises indefinitely
- D. Repurchase a SaaS core banking product
Correct answer: B
Explanation: The deadline is the deciding constraint. Rehosting is the fastest low-risk route, and RDS takes away database administration work. Modernising afterwards cuts licensing costs gradually, with each step tested and reversible, which is good practice for banking change management. Babelfish compatibility must be assessed first, because not all T-SQL features are supported.
Why the others are wrong:
- A: A full rewrite won't be finished in 4 months, so it misses the exit deadline and carries high risk.
- C: Retaining the app goes against the data-centre exit.
- D: Repurchasing is a business decision far bigger than the brief, and it can't be done in 4 months.
Independent learning resource — not affiliated with, endorsed or sponsored by Amazon Web Services. AWS and related marks are trademarks of Amazon.com, Inc. or its affiliates. All banking data in examples is fictional. Sample architectures are for learning and are not automatically compliant with GDPR, PCI DSS or any banking regulation.
AWS Certified DevOps Engineer – Professional (DOP-C02)
The professional DevOps exam. It covers CI/CD, infrastructure as code, resilience, monitoring, incident response and compliance automation.
AWS Certified AI Practitioner (AIF-C01)
Optional foundational exam: AI, ML and generative AI concepts, foundation model applications, responsible AI and AI governance on AWS.