A genomics lab runs nightly variant-calling pipelines for 6 hours on 200 vCPUs with sustained CPU pressure, plus event-driven image validations that run 400 ms 5 million times monthly. The budget was cut 25% and performance must stay constant while reducing compute and orchestration spend. What cost-optimized design helps?

The correct design matches each workload pattern to its lowest-cost engine while holding performance constant. Graviton-based instances deliver superior price-performance for sustained CPU-bound variant calling over six hours on 200 vCPUs, lowering per-core cost versus comparable x86, Lambda provides scale-to-zero per-millisecond billing ideal for event-driven 400 ms validations occurring five million times monthly, and Express Workflows provides low-cost orchestration optimized for high-volume short-lived executions. The Graviton plus Lambda with Standard Workflows approach fails on orchestration economics because Standard bills per state transition for durable long-running executions with full history, which becomes prohibitively expensive at this volume where Express bills by executions and duration. The x86 with Savings Plans plus Lambda with Express approach fails on processor efficiency because a commitment discount lowers the rate but retains the higher baseline cost and misses the architectural gain required to absorb a 25 percent cut under sustained pressure. The Graviton plus Fargate with Express approach fails on granularity because Fargate bills for provisioned vCPU and memory with startup overhead suited to longer containers, wasting idle spend on bursty sub-second work where Lambda eliminates provisioning. Exam caveat is to map execution count and duration to workflow type rather than defaulting to durable orchestration. Practical check is to benchmark the pipeline on Graviton for compatibility, load-test validation concurrency, and model monthly Lambda plus Express charges against current spend.