Railway bills purely on actual resource consumption (CPU and memory metered by the minute), not on reserved capacity, so nothing is "reserved" for you and there is no flat fee per replica. With 50 replicas you pay for whatever CPU and memory all 50 instances actually consume, summed together, so even idle replicas accrue cost for their baseline idle memory footprint multiplied across all 50. Resource rates are $10/GB/month for RAM and $20/vCPU/month for CPU, detailed in our pricing docs, and each replica can independently use up to your plan's per-service limits as described in the scaling docs.

[{"text": "Railway bills on actual resource consumption metered by the minute, not reserved capacity, with no flat per-replica fee", "sourceToolCallIds": ["searchDocs"]}, {"text": "You pay for the summed CPU and memory all replicas actually consume", "sourceToolCallIds": ["searchDocs"]}, {"text": "RAM is $10/GB/month and CPU is $20/vCPU/month", "sourceToolCallIds": ["searchDocs"]}, {"text": "Each replica can independently use up to the plan's per-service resource limits", "sourceToolCallIds": ["searchDocs"]}]