ECS and EKS cost forecasting for engineering leads

The moment an engineering lead or founder realizes they have this problem is when an AWS bill arrives substantially higher than expected, and they cannot tell whether it was the Fargate tasks, the load balancer hours, the NAT gateway traffic, or the EKS control plane fee that caused the spike — because AWS Cost Explorer shows individual line items but not the total cost of a logical service or deployment.

The gap is structural. AWS bills by resource type, not by application or team. A single containerized service touches Fargate compute, ECR storage, a load balancer, a NAT gateway, CloudWatch logs, and potentially Secrets Manager — all billed separately with no native grouping by 'this is the cost of running my payments service.' Users say 'costs can add up, especially when scaling or using additional services,' but the problem is that they can't see the accumulation happening until after the fact.

The specific issue is that choosing between EC2 and Fargate for sustained workloads requires modeling reserved instance pricing, spot interruption risk, and Fargate's per-vCPU-per-second rates simultaneously — most engineers make this decision once, get it wrong, and don't revisit it until the cost is already a problem. There's no native AWS tool that shows: 'at your current workload profile, switching this cluster to a mix of reserved EC2 and Fargate spot would save $X/month.'

This is a business because cost decisions in containerized infrastructure recur every time a workload scales, every time a new service is deployed, and every time a team is asked to cut the cloud bill. The buyer (engineering lead, CTO at a startup, cloud finance role) is different from the daily user (developer), which is exactly why AWS doesn't fix it — developers don't control the budget conversation, and budget holders don't live in the console.

What to build

Build a cost attribution layer that reads AWS Cost and Usage Reports and CloudWatch metrics, groups spending by ECS service or EKS namespace, and generates a monthly forecast showing what each logical service actually costs end-to-end — plus a what-if comparison between Fargate, EC2 reserved, and Spot configurations at current usage patterns.

Where to start

Target YC and similar batch companies six to twelve months post-launch when their AWS bill has just crossed $5K/month and they're being asked to explain it — this is the exact moment they search for cost visibility and have no internal tooling yet.

The hard part

The hardest early challenge is getting enough real CUR data from design partners to validate that the attribution logic handles edge cases correctly — tagging inconsistency across AWS accounts is endemic, and a cost report that's confidently wrong is worse than no report at all.

How it makes money

Percentage of cloud spend under management, starting at 1% with a monthly minimum of $99, which aligns incentives — customers who save money keep paying, and growth in their cloud spend grows revenue naturally.