Techniques, thinking patterns and a playbook, distilled from 28 public, numbers-backed accounts of AWS cost reduction published between 2012 and September 2026: Levels.fyi, Segment, Duolingo, Lyft, Airbnb, Canva, Snap, Rewind, Pinterest, Honeycomb, Zomato, Zendesk, Salesforce, Tinybird, Moneysmart, HERE, Amazon Prime Video, Recall.ai, DoorDash, nOps, 37signals, Dropbox, X, Prerender.io, OneUptime and GEICO. The series explains how the engineers reasoned, what they refused to trust and what they got wrong. Savings are company-reported and unaudited; treat every number as a signal of what is possible, then measure your own.
Download the full research report. Cutting the AWS Bill: 28 Case Studies, 53 pages, PDF, about 780 KB. One click starts the download; the file also lives in this site's GitHub repository.
⬇ Download PDFStart with Part 1 for the vocabulary, then jump to the lever that matches your biggest line item.
New: the seven billing meters, Mumbai list prices, Cost Explorer usage types, a formula, trap and lever for firewall, NAT, load balancers, EKS, EC2, EBS, S3, CloudWatch and more, plus how to estimate a component before you build it.
The cost equation (quantity x unit price x time x topology), twelve levers, an eight-step order of operations, four shapes of spend, fourteen thinking patterns and the scoreboard of all 28 cases.
Levels.fyi (twice), Segment, Duolingo, Lyft and Airbnb: turn billing data into something queryable, question every large number, and verify from the inside before deleting.
Canva, Snap and Rewind: bytes pay the bill, objects pay the migration. The S3 break-even formula, lifecycle rules and a $103K invoice with a six-month payback.
Pinterest, Honeycomb, Zomato and Zendesk: 30-47% lower compute cost from a CPU-architecture change, with a pilot plan, benchmark method and Kubernetes rollout.
Salesforce (twice), Tinybird and Moneysmart: Spot, Karpenter consolidation and off-hours schedules, and the workload discipline that makes them safe.
HERE's $50M commitment strategy plus Prime Video, Recall.ai, DoorDash and nOps: where the invisible costs hide in per-transition billing, IPC copies, cross-AZ chatter and one NAT route.
37signals, Dropbox, X, Prerender.io, OneUptime and GEICO: when leaving pays, when it does not, and a total-cost-of-ownership skeleton to test it.
Every technique by AWS service, the eight reusable formulas and twenty pitfalls with a guard rail each: a checklist for any cost review.
A 90-day playbook, the operating rhythm and KPIs that stop savings eroding, and concrete checks for an EKS, Vault, Argo CD, pgBackRest and SQL Server platform.
25 collapsible, one-click-copy prompts for Claude, Claude Code or any AI agent: read-only guard rails, discovery, lever audits, verification, a lead-specialist-critic subagent pattern, and a downloadable Markdown library.