Microseconds on one side, an audit trail on the other.
Two workloads, two completely different network designs, deliberately not shared.
Trading engine
Cluster placement group + NLB — every non-essential hop removed for minimum latency
Customer-data platform
Separate VPC, no direct route, Network Firewall logs every crossing for audit
Wildly different requirements, raw speed vs. airtight segmentation, are exactly why they get architecturally separate treatment.