Three genuinely different workload shapes, three platforms.
Meridian's own three-way split proves the decision table rather than being arbitrary.
shipment-api
Cloud Run
Spiky, customer-facing HTTP
route-optimizer
GKE
Needs closer runtime control as ML deps grow
GPS-ingestion
Compute Engine
Steady, interruption-tolerant
The worked cost comparison confirms the crossover: GKE Autopilot wins below ~60-70% sustained utilization, Standard wins above it, Cloud Run wins for a genuinely spiky, mostly-idle profile.