Container (ECS/EKS) vs Serverless (Lambda) cost

Lambda or ECS — the volume crossover

Lambda is cheap at low volumes (often free) and expensive at high volumes. Containers are expensive at low volumes (always-on bill) and cheap at high volumes. The break-even depends on request rate, duration, and memory.

The math

GB-seconds per request = memory MB ÷ 1024 × duration ms ÷ 1000
total GB-seconds = requests × GB-seconds per request
Lambda cost = total GB-seconds × $0.0000166667 + requests × $0.0000002

Default scenario: 50M requests/mo × 250ms × 512MB:

• GB-seconds: 50M × 0.125 = 6.25M
• Lambda cost: 6.25M × $0.0000166667 + 50M × $0.0000002 = $104 + $10 = $114/mo
• Container cost (1 small instance): $320/mo
• Lambda wins by $206/mo at this volume.

When containers win

• Sustained throughput → instance is "fully utilized" anyway
• Long-running connections (websockets, gRPC streams)
• Latency-sensitive (Lambda cold start adds 100-500ms first call)
• High memory (Lambda max 10GB; certain workloads need 32+ GB)
• Workloads needing GPU

When Lambda wins

• Bursty / unpredictable traffic
• Long idle periods
• Sub-second functions
• Event-driven (S3 → Lambda, SQS → Lambda)
• Cost optimization at low scale

The break-even

Default scenario: $320 / per-request cost ($0.00000228) = ~140M requests/mo before Lambda exceeds container cost. Above 140M req/mo: containers cheaper. Below: Lambda cheaper.

What this calc misses

• NAT Gateway / VPC costs (both apply differently)
• Cold start latency (real Lambda cost in user experience for low-traffic functions)
• Reserved instance discounts on containers (-40-60%)
• Lambda Provisioned Concurrency (eliminates cold start; costs more)
• API Gateway costs ($1.00 per million requests for REST API on top of Lambda)