Migrating AI Workloads from Azure to AWS
Inquizyt is an innovative technology company delivering cutting-edge services powered by large language models (LLMs). Their production platform was originally running GPU-based inference workloads on Microsoft Azure. As the company scaled, Inquizyt looked to AWS to achieve greater cost efficiency, cloud-native scalability, and long-term modernization, while also benefiting from AWS’s industry-leading customer support.
Business & Technical Goals
The customer wanted to reduce the total cost of ownership and improve scalability by moving GPU-intensive LLM workloads from Azure to AWS. Key objectives included:
- Optimizing GPU usage with right-sized instances and cost savings plans
- Deploying a secure, scalable, and maintainable AWS environment
- Refining Infrastructure as Code and DevOps practices
- Building flexibility for future modernization, including AWS Bedrock
Migration Readiness & Discovery
We began with a Migration Readiness Assessment (MRA), including stakeholder workshops and a Cloud Readiness Assessment. This defined the customer’s cloud maturity level, potential risks, and actionable improvements.
During Discovery, we manually reviewed the Azure infrastructure, workloads, and pipelines, and evaluated the existing AWS account. Findings were documented in a Portfolio Discovery report, capturing systems, dependencies, and migration considerations.
TCO & Planning
Next, we created a TCO estimate using the AWS Pricing Calculator, comparing Azure workloads with the proposed AWS architecture (EC2, GPU instances, RDS, ALB, ECR, networking). This highlighted potential cost savings and scalability benefits.
Based on these insights, we built a Migration Project Plan with clear phases, milestones, and responsibilities. The plan defined the target AWS Landing Zone, architecture, and communication framework between customer and Noveo teams.
Building the Foundation
We designed and deployed a Landing Zone following AWS Well-Architected Framework principles. This included:
- Multi-account structure with AWS Organizations
- IAM Identity Center with SSO and permission sets
- Security baseline (SCPs, KMS, MFA)
- Multi-AZ VPCs with secure networking
This provided a secure and scalable foundation for all future workloads.
Pilot Migration (POC)
To validate the approach, we deployed a Demo environment using reusable Terraform modules. The setup included EC2 with Auto Scaling, RDS, ALB, Secrets Manager, IAM roles, and KMS keys.
We integrated with the customer’s Azure DevOps pipelines, pushing Docker images to ECR and deploying via SSM Run Command. The pilot confirmed architecture design choices and created a reusable blueprint for the full migration.
Reskilling & Handover
To ensure long-term independence, we delivered reskilling workshops covering AWS setup, Landing Zone structure, and hands-on onboarding for application deployment.
The team received full documentation and curated learning resources, enabling them to manage and extend AWS environments independently after project handover.
Results
- Reduced TCO through optimized GPU infrastructure and AWS pricing models
- Secure and scalable AWS foundation aligned with best practices
- Validated migration approach via pilot environment and reusable Terraform modules
- Empowered customer team with training, documentation, and autonomy
This migration gave the customer a modern AWS environment ready for AI workloads, with clear cost benefits and long-term scalability.