GCP to AWS Migration Assessment

Migration Decision Summary

Go, with conditions

Execution shape: Phased Migration · Complexity: Complex

By track: Compute + database: go. AI text: go. AI image: conditional — keeps the current provider via the adapter until the quality eval passes; does not gate the rest. Analytics: deferred — specialist track, parallel.

• Condition: confirm database availability — Multi-AZ was assumed, not confirmed (roughly 2x database cost factor)
• Condition: confirm CUD expiration date before committing a migration start date

Confidence: Medium — full basis in What This Assessment Rests On.

Migrate if

• You want unified AWS operations for containers, database, and Bedrock AI
• Estimated combined cloud-service savings (~$763/month) and AWS Activate credits offset the cost of migrating
• You can engage AWS account team for BigQuery analytics in parallel

Stay entirely if

• You require BigQuery analytics in the same cutover window as app infra and cannot run a phased analytics track with specialist planning
• GCP Committed Use Discounts (CUDs) or team GCP expertise outweigh savings
• Cost is the only metric and a dual-run period must be avoided

What would flip this

• Single-AZ acceptable: AWS estimate drops further, strengthens go
• BigQuery must cut over in the same window: defer for specialist evidence

Key decisions ahead

1. Target region and deployment model (Fargate vs EKS)
2. Database migration tooling and cutover window
3. Bedrock model selection for gemini-2.5-flash and imagen-3.0-generate-001
4. BigQuery analytics path with AWS account team or data partner

Next steps

1. Review financial case and security posture with stakeholders
2. Engage AWS account team for BigQuery analytics migration path
3. Confirm Bedrock model access and quota in us-west-2
4. Apply for AWS Activate Founders credits before ramping usage
5. Schedule phased cutover per cluster creation order

Share With Your CFO

The one-paragraph version, ready to forward:

Estimated AWS Monthly Run Rate

Combined view of recurring infrastructure and AI cloud-service costs. BigQuery is excluded from both sides. This is not a total ownership-cost model: staffing, operations, support, and migration labor are not priced.

* Optimized AI assumes Nova Lite for text + Stable Image Core for images after A/B validation (see the optimized AI projection in Appendix D).

GCP infrastructure breakdown (baseline)

Current Stack Overview

5 primary GCP services detected:

• Cloud Run (compute) — sf-beach-ratings
• Cloud SQL PostgreSQL (database) — sf-beach-db
• Cloud Storage (storage) — beach images + cloudbuild
• BigQuery (analytics) — sf_beach_analytics (deferred)
• Vertex AI (AI) — Gemini + Imagen via application code

Infrastructure Cost Comparison

Three pricing scenarios for the same mapped architecture. The generated terraform/ implements the Balanced tier. The full per-service breakdown is in Appendix B.

Cost Optimization

Generated Terraform is Balanced on-demand (Est. $112/mo). Savings Plans and Reserved Instances are account-level purchases after you have a usage baseline. Do not add these rows on top of the Optimized tier (Est. $82/mo) — that tier already embeds illustrative Spot and 1-year Savings Plan assumptions.

Line-level targets are in Savings Plans and Reserved Instances.

Architecture Overview

High-level service mapping

Architecture diagram

Migration cluster order

1. networking_vpc_us-west1_001 — firewall rules → security groups
2. storage_gcs_us-west1_001 — beach images bucket
3. storage_gcs_us_001 — Cloud Build artifacts
4. database_sql_us-west1_001 — Cloud SQL → RDS
5. compute_cloudrun_us-west1_001 — Cloud Run → Fargate
6. Deferred: analytics_bigquery_us_001

Security Posture

The generated terraform/baseline.tf deploys an account-wide Tier 1 security baseline — CloudTrail audit logging, GuardDuty threat detection, IAM password policy, S3 public-access block, EBS default encryption, and AWS Budgets — for a $15/mo mid estimate (range $3–$30, ±25%). GuardDuty is the dominant cost at $13/mo after its 30-day free trial.

Full control-by-control detail, GCP equivalents, and the GCP→AWS gap analysis are in Appendix E and Appendix F.

Migration Shape

Infrastructure migration (phased, in dependency order)

AI migration (parallel track)

Duration drivers: database replication and cutover (binding), cross-cluster security groups, parallel AI evaluation per model. Stage order and go/no-go gates carry the schedule; the plugin quotes no week or hour estimates (uncalibrated).

Cutover strategy: maintenance-window-weekly  |  Rollback: revert DNS to GCP Cloud Run; keep RDS read replica 48h

Top Risks

What This Assessment Rests On

Assumptions applied by default — settings you never confirmed; each one shifts the numbers as described:

Confidence: Medium — some material inputs assumed. Measured: GCP billing baseline ($165/mo infra), Cloud SQL disk size (10 GB). Assumed: database availability, Cloud Run traffic pattern. Unknown: none.

Pricing provenance: Estimates based on cached AWS pricing (2026-03-07), accuracy ±5–10%. Cross-checked against the AWS Pricing Calculator for the Balanced scenario.

Assumptions, Exclusions & Validation

Pricing confidence

Exclusions

• BigQuery analytics — AWS target path not yet selected by specialist; costs excluded from the estimated AWS monthly run rate (does not block other services)
• GCP egress during migration — billing export not available (billing_data_available: false)
• Professional services / labor — neither estimated nor priced (no calibrated effort data)
• Dual-run period — temporary cost of running GCP + AWS concurrently not modeled
• Dev-tier sizing — db.t4g.micro, single-AZ, on-demand Fargate unless upgraded

Terraform validation

Design warnings

• Architecture validation unavailable (AWS MCP Server not invoked — non-critical)
• Public S3 access for beach images should use CloudFront in production

Appendix A — Service Recommendations

Appendix B — Cost Estimates

Per-service breakdown (AWS Balanced tier)

Tier totals (Premium / Balanced / Optimized) and annual differences are in the Infrastructure Cost Comparison above. Commitment discounts are in Savings Plans and Reserved Instances.

Non-cost benefits (from ROI analysis)

• Reduced operational overhead from managed Fargate and RDS
• AWS-managed patching and scaling for database and containers
• Unified AWS billing and IAM for infra + Bedrock
• 30+ Bedrock models without multi-vendor API keys; Guardrails and Knowledge Bases for future Retrieval-Augmented Generation (RAG)
• Vendor diversification away from single-cloud GCP dependency

Savings Plans and Reserved Instances

Incremental to Balanced on-demand (Est. $112/mo). Do not add these savings onto Optimized (Est. $82/mo) — that tier already embeds reservation and Spot assumptions.

Appendix C — Migration Steps & Rollback

Phase 1 — Prerequisites

• AWS account with billing alerts; enable root MFA; restrict root to break-glass use
• Apply for AWS Activate Founders ($5,000) — see STARTUP_PROGRAMS.md
• Run scripts/01-validate-prerequisites.sh --execute

Phase 2 — Infrastructure (Terraform)

cd terraform
terraform init -backend=false
terraform apply -target=aws_s3_bucket.tfstate -target=aws_dynamodb_table.tfstate_lock
terraform init
terraform plan
terraform apply

Deploys Balanced-tier stack including baseline.tf security controls.

Phase 3 — Data migration

• ./scripts/02-migrate-data.sh --execute — GCS → S3, DMS for PostgreSQL
• ./scripts/03-migrate-containers.sh --execute — GCR → ECR

Phase 4 — Cutover & validation

• Weekly maintenance window cutover per generation-infra.json
• ./scripts/05-validate-migration.sh --execute
• Keep GCP hot standby 48 hours

Rollback procedure

Appendix D — AI Migration

Model mappings

Model cost comparison

Recommended model token breakdown (Claude Sonnet 5)

Code changes

Production readiness checklist

• Bedrock model access enabled for Claude Sonnet 5 and Stable Image Core in us-west-2
• IAM role with bedrock:InvokeModel on ECS task role
• Provider adapter deployed with AI_PROVIDER feature flag
• A/B test with ≥100 representative prompts; image quality human evaluation
• Monitoring: request volume, P50/P95 latency, token usage, daily cost
• Rollback procedure tested

Success criteria

• Response quality ≥ 90% of Gemini baseline
• P95 latency within 2× of Gemini baseline
• Monthly cost within 20% of estimation-ai.json projection

Your Configuration Choices

Questions and assumptions from the Clarify phase that shaped this migration plan.

Appendix E — AWS Security Posture (Tier 1 Baseline)

The generated terraform/baseline.tf implements an account-wide security baseline. Costs below from estimation-infra.json → projected_costs.breakdown.security_baseline (Balanced mid estimate: $15/mo, range $3–$30, accuracy ±25%).

Application-layer security (security.tf)

Appendix F — GCP → AWS Security Gap Analysis

Appendix G — Generated Artifacts Catalog

Appendix H — Glossary