Switch to Hugging Face Inference Endpoints for ML inference — migrate from ECS/Fargate

AI Impact Summary

An organization is migrating inference workloads from AWS ECS/Fargate to Hugging Face Inference Endpoints, leveraging Hugging Face Hub as the model registry. Benchmark tests on a RoBERTa-based text classification model show lower latency on Inference Endpoints than the previous ECS deployment, highlighting performance benefits for real-time inference. The shift reduces deployment overhead and keeps models tightly integrated with Hugging Face tooling, but it increases ongoing endpoint costs by roughly 24-50% per endpoint, demanding budget reallocation if scaling to many models.