We increased our Node service's parallelism by 30 times, which equated to a cost savings of about $300k annually. To achieve this, we carefully planned and executed the rollout, relying on good tooling, observability, and debugging. We addressed several bottlenecks in our system, including memory allocation issues due to V8 heap size limitations, garbage collection time optimization, CPU usage optimization, and JSON serialization performance improvements. Our approach involved using Prometheus metrics, Grafana dashboards, LaunchDarkly feature flags, 0x package for profiling, and local benchmarks to identify and fix performance issues. The migration simplified our architecture and reduced our annual EC2 instance spend by $300k while making our system more robust to external request latencies or spikes in API traffic from customers.