The Problem
Early-stage biotech startups often over-provision AWS infrastructure as a precaution, then watch costs compound as the team grows faster than the infrastructure is audited. ImmunisAI had accumulated orphaned EC2 instances from abandoned experiments, over-sized RDS clusters that were provisioned for peak loads that never materialized, and multi-region redundancy configurations that served no production purpose. Their monthly AWS bill had tripled in 12 months without proportional growth in traffic, users, or workloads. This is a pattern common in fast-moving teams where engineering bandwidth goes entirely to product development and infrastructure hygiene falls off. Leadership knew they were overspending but lacked the expertise to identify exactly where and to optimize without risking downtime or data loss on live research workflows.
How We Solved It
We started with a full spend audit using AWS Cost Explorer, Trusted Advisor, and Compute Optimizer, mapping every dollar of monthly spend to its originating resource and usage pattern. The analysis identified the specific waste categories and sized the savings potential before a single change was made. From there, we executed a systematic optimization over two months: right-sizing 8 EC2 instances to more appropriate instance families and sizes based on actual CPU and memory utilization data; converting 3 on-demand RDS instances to 1-year reserved capacity at significantly lower effective hourly rates; eliminating 4 idle Elastic IP addresses and the associated NAT gateway charges they were generating; and collapsing an unnecessary multi-region architecture that was generating avoidable cross-region data transfer costs with no redundancy value. All changes were executed live using blue/green swap patterns and RDS parameter group updates. No maintenance windows, no planned downtime, no risk to research data or active workloads throughout the entire two-month optimization cycle.
What We Delivered
- 30% reduction in monthly AWS costs
- Completed in just 2 months
- Zero downtime during optimization