Newcast is a marketing enablement platform that uses artificial intelligence to generate compelling, high-quality videos from product descriptions, Shopify, or Amazon links. With a strong focus on cost-efficiency and rapid delivery, Newcast empowers brands to scale video content creation at unprecedented speed.

Operating on another cloud provider, Newcast faced a series of performance and operational inefficiencies that hindered their ability to scale effectively. Their inference workloads, which relied on GPU instances, were not optimized for bursty usage and lacked the scalability needed to support growing demand. The infrastructure depended heavily on standalone scripts and tools like FFMPEG, limiting flexibility and slowing down deployment cycles. Additionally, critical components such as Jupyter Notebooks, bash-based automation, and the monolithic ComfyUI pipeline required modernization and more efficient orchestration. These constraints highlighted the need for improved management of containerized applications and a more scalable, future-proof infrastructure, prompting Newcast to pursue a strategic migration to a cloud environment that could better support their technical and business goals.

AWS was selected as the preferred cloud provider due to its comprehensive support for GPU-optimized instance types such as G5, G6, and G6e, which offered Newcast the flexibility to right-size their workloads efficiently. Its advanced container orchestration capabilities through Amazon EKS, combined with Karpenter for intelligent auto-scaling, provided a scalable foundation for their AI-driven infrastructure. Additionally, AWS’s native compatibility with Terraform enabled infrastructure as code, ensuring repeatable and consistent deployments. The availability of integrated managed services—including Amazon S3, AWS Elemental MediaConvert, and Amazon SageMaker—further aligned with Newcast’s modernization goals, supporting both current needs and future innovation.

SnapSoft executed a phased Migration Acceleration Program (MAP) to transition Newcast’s infrastructure to AWS, starting with a comprehensive assessment and concluding with full mobilization. During the MAP Assess phase, SnapSoft conducted a detailed inventory of Newcast’s workloads, including inference workloads running on L4 GPUs, AI pipelines utilizing Stability AI and LoRA, FFMPEG processing, Jupyter Notebooks, and the ComfyUI pipeline. They also evaluated modernization opportunities, such as replacing manual video processing with AWS Elemental MediaConvert and implementing serverless architectures for automation. In the MAP Mobilize phase, SnapSoft migrated GPU workloads from existing instances to AWS instances (g6.12xlarge and g6.4xlarge), leveraging Amazon EKS with Karpenter for Kubernetes orchestration. The entire infrastructure was defined and deployed using Terraform, encompassing GPU nodes, S3 buckets for model, input, and output storage, a Docker repository via ECR, and the deployment of ComfyUI on GPU-optimized environments. GitHub Actions was configured to automate CI/CD processes, while SQS queues and VPC endpoints enabled scalable, event-driven processing. Integration with NoCode and LowCode environments further enhanced flexibility. This solution significantly improved manageability, scalability, and deployment consistency through a modular, decoupled architecture.

• Amazon Elastic Kubernetes Service (EKS)
• Amazon EC2 (g5, g6, g6e instances)
• Amazon S3
• AWS Elemental MediaConvert
• AWS Lambda
• Amazon SQS
• AWS Systems Manager (SSM)
• AWS VPC & Networking
• Amazon SageMaker (planned)
• Terraform (Infrastructure as Code)
• GitHub Actions (CI/CD)
• Docker & Amazon ECR

Performance Gains

• High-throughput GPU instances improved inference times.
• Container orchestration optimized resource usage via Karpenter.

Cost Optimization

• Right-sized instance selection reduced over-provisioning.
• AWS funding fully offset engineering costs (~$107,980 Mobilize phase).

Operational Efficiency

• Infrastructure-as-code (Terraform) ensured repeatable deployments.
• GitHub Actions automated container builds and deployments.
• Setup of decoupled pipelines improved development velocity.

Scalability and Future-readiness

• EKS architecture and spot instances prepare the environment for scale.
• Modular infrastructure supports experimentation with AWS managed AI services.