I scaled an IoT platform to 2M devices on Lambda. Today, I'd run the hot path on containers.
Serverless was the right call early. Traffic was spiky, unpredictable.
Then the pattern changed. Device telemetry ingestion hit constant high throughput, 24/7. Lambda was millions millions of invocations on a workload that never scaled to zero. We cut log volume, trimmed memory, batched SQS consumers. Hit the floor.
That hot path on ECS Fargate. (Almost) same code, containerized, sustained compute. The rest of the platform stayed on Lambda where it belonged.
Serverless-first doesn't mean lambda-forever. It means Lambda is your default until the invocation graph says otherwise. Flat line? You're overpaying.
Have you re-evaluated a foundational architecture decision mid-project? What triggered it?