Cloud spend is growing faster than revenue. As Cloud Architect, how do you cut 30% without hurting reliability?
Short answer
Get visibility by team and unit cost, remove waste first (idle, oversized, non-prod schedules, storage and log retention, data transfer), buy commitments for the stable baseline, then apply architectural levers (ARM instances, Spot, caching, serverless for spiky loads), and institutionalize budgets and anomaly alerts.
Detailed answer
| Step | Actions | Typical impact |
|---|---|---|
| Visibility | Tag compliance, allocation, unit cost dashboard | Enables decisions |
| Waste | Idle resources, orphaned volumes/snapshots, non-prod off-hours, oversized instances | 10–20% |
| Commitments | Savings Plans/CUDs/reservations on baseline | 10–25% of covered spend |
| Architecture | Graviton/ARM, Spot for stateless, caching, CDN, VPC endpoints to cut NAT, log volume controls | 5–20% |
| Operate | Budgets, anomaly detection, cost in design reviews | Sustains savings |
Protect reliability: never cut redundancy for Tier 0/1 without explicit risk acceptance; test rightsizing under load.
What the interviewer expects
Sequenced approach, specific levers, unit economics, reliability awareness.
Common mistakes to avoid
Buying three-year commitments before rightsizing; cutting DR to save money without business agreement.