The Hidden AWS Discount: When Spot and Graviton Are Safe to Use
A team stopped cutting AWS resources and started questioning how they paid for them, unlocking Fargate Spot and Graviton savings while learning why some costs like RDS Proxy must stay.
When a team's AWS bill crept over budget two months running, they found their resources were already lean - logs trimmed, tracing sampled, instances right-sized. The real opportunity wasn't usage but purchasing method. They moved their staging API to Fargate Spot, AWS's roughly 70% cheaper channel for idle capacity that can be reclaimed within two minutes - fine for staging with no real users, unacceptable for production.
The second move was switching Fargate tasks to Graviton, AWS's own ARM chip, cutting costs about 20% with no availability trade-off, since the savings come from AWS eliminating the Intel/AMD margin and running cooler, cheaper hardware. Because Graviton carries no reclamation risk, it went straight to production too, with only cold-start latency worth monitoring.
The flip side surfaced when RDS Proxy, costing $22/month, looked like a wasteful line item since its rotation feature was disabled. In fact its real job was preventing connection-surge failures during deploys - cutting it would have removed a safety net, not saved money. The team fixed a misleading code comment instead of cutting the resource.
The broader lesson: before trimming any cloud cost, ask why the number exists. Some costs are unclaimed discounts (Spot, Graviton) with knowable trade-offs; others are insurance against real incidents (like RDS Proxy) that shouldn't be touched at all.