AssemblyAI cuts speech-to-text inference costs 70% on AWS with ECS and Spot GPUs

AWS Machine Learning described how AssemblyAI runs its speech-to-text inference pipeline on Amazon ECS, EC2 Spot Instances and Amazon SQS. The architecture uses a flexible fleet of NVIDIA GPUs with ECS capacity providers, mixing Spot and On-Demand capacity across multiple instance families and regions. AssemblyAI keeps independent stacks in US West (Oregon) and Europe (London) to meet data residency requirements. The company said the approach cut inference compute costs by 70% while maintaining SLA commitments across European AWS Regions.