AWS Cost Optimization & FinOps
Writes the FactualMinds guides on AWS cost optimization and FinOps: pricing models, commitment planning, and keeping cloud spend predictable.
Checked 10 October 2026: managed Bedrock knowledge base storage is $5 per GB of raw data a month. One thousand GB is $5,000 before retrieval. An e-commerce workspace of terabytes of Drive files and millions of Gmail messages reached hundreds of dollars a day after ingestion started.
A 5,000-document corpus on the site calculator is $163.28/month on S3 Vectors and $510.52/month on OpenSearch Serverless Classic. $160 of the smaller bill is Claude Sonnet 5.5.
AWS Settings spend limits, still a limited release on 23 Sep 2026, alert at 50%, 75%, and 90%, then pause the project. Do nothing for 90 days and AWS deletes the project data. The minimum limit is the greater of $20 or AWS's estimate.
On Aug 6, 2026 ElastiCache added Graviton4 M8g/R8g/C8gn for Valkey and Memcached — up to 47% higher throughput, 43% lower P99, 31% better price-performance vs Graviton3. Field guide: family pick, RI traps, canary checklist.
On July 16, 2026 AWS removed the 30-day Standard residency gate for Standard-IA and One Zone-IA lifecycle transitions — day-0 IA can cut first-month storage ~46% on 10 TB cold logs, but the 30-day IA billing minimum and $0.01/GB retrieval fees still apply.
CloudZero, Vantage, Finout, nOps, ProsperOps, and Kubecost on AWS — platform selection guide plus who implements tagging, allocation, and architecture savings.
Production guide for Kubecost on AWS EKS — cost allocation setup plus architecture changes that reduce spend, not just attribute it.
Implement ProsperOps on AWS — Savings Plans automation works best after baseline modeling and architecture stability. Production checklist included.
EC2 On-Demand is a matrix of family, size, OS, tenancy, and region. In us-east-1 (June 2026) a m7g.large Linux runs $0.0816/hr vs a g5.xlarge GPU at $1.006/hr. Graviton5 M9g/C9g now GA for new fleets.
ElastiCache bills node-hours for Redis/Valkey clusters and ECPUs for Serverless. A cache.r7g.large with two replicas is 3× the primary node cost before data transfer. Serverless wins below ~$800/mo steady cache spend; provisioned nodes win above it.
Fargate bills per vCPU-second and GB-second, not per task hour. In us-east-1 a 0.25 vCPU / 0.5 GB task runs ~$9/mo; a 4 vCPU / 16 GB task ~$145/mo. Prefer arm64 + C9g/C8g when moving steady-state to ECS on EC2.
A 22-account AWS Organization spent $1.1M/yr on Compute Savings Plans but applied only 61% to production — dev sandboxes burned the commit while prod stayed On-Demand. Group Sharing (April 2026) fixed attribution; the waste was $312k/yr before they restructured purchases.
Our articles share what we know. Our consulting engagements apply that knowledge to your specific AWS environment.
We use cookies and similar technologies to analyze site traffic, personalize content, and provide social media features. By clicking “Accept,” you consent to our use of cookies. You can adjust your preferences at any time.
