Cost Check

Enterprise optimizes for: lowest per-unit cost at scale. Startups optimize for: lowest BASELINE cost with LINEAR scaling (no cliffs).

The Three Cost Questions for Startups

1. What’s my cost at zero/minimal traffic? (This is your monthly burn before you have customers)
2. What’s my cost at 10x current traffic? (Is it linear or does it cliff?)
3. What’s my cost when credits expire? (The real number)

Hidden Cost Killers — Startup-Specific

These rarely appear in enterprise cost reviews but destroy startup budgets:

Credits Strategy

How to Model Costs with Credits

Credits Optimization Rules

1. Never buy Reserved Instances or Savings Plans while on credits — you can’t see real usage patterns
2. Credits cover everything except Marketplace — use native AWS services, not Marketplace alternatives
3. Track credits burn rate monthly — if burning faster than expected, investigate NOW not at expiry
4. Plan architecture for post-credits reality — if your arch costs $15K/mo and revenue is $5K/mo, you have a problem BEFORE credits expire
5. Credits don’t carry over after expiry date — use-it-or-lose-it. Don’t under-utilize to “save” them

Cost Scaling Patterns — Choose Wisely

The “$100/mo Baseline” Rule

If your architecture costs >$100/mo at zero customers, justify every dollar:

• $16/mo ALB: Do you need it, or can Lambda + API Gateway work?
• $32/mo NAT Gateway: Do you actually need private subnet egress?
• $43/mo Aurora Serverless v2 minimum: Is RDS t4g.micro ($12/mo) sufficient?
• $7/mo per VPC endpoint: Can you use public endpoints with IAM auth instead?

When to Invest in Cost Optimization

Quick Wins Checklist

• Not done: Unused EBS volumes and unattached Elastic IPs
• Not done: CloudWatch log retention set to “Never expire” — change to 7d dev / 30d prod
• Not done: NAT Gateway traffic that could use VPC endpoints
• Not done: Over-provisioned RDS instances (check CPU utilization)
• Not done: Lambda functions with excessive memory allocation
• Not done: Dev environments running 24/7 — schedule stop outside business hours
• Not done: Old EBS snapshots and unused AMIs
• Not done: S3 buckets without lifecycle policies

Gotchas

• Data transfer costs are the silent killer — especially cross-AZ and cross-region
• DynamoDB on-demand vs provisioned: on-demand is cheaper below ~20% utilization of provisioned capacity
• S3 Intelligent-Tiering monitoring fee per object — not worth it for millions of tiny objects
• CloudFront can be cheaper than S3 direct for high-traffic reads (no S3 request fees)
• Graviton instances are ~20% cheaper and often faster — use them unless you need x86