Compute Savings Plans

TL;DR: A Compute Savings Plan is an all-access discount pass: commit to a steady $/hour of compute for 1 or 3 years and get up to 66% off on-demand — and the discount applies automatically across EC2, Lambda, and Fargate, regardless of instance family, size, OS, or region. Unlike Reserved Instances (a specific treadmill at a specific gym), the discount follows your workload as it evolves — EC2 today, Lambda tomorrow, a different region next month. For most teams whose architecture changes, that flexibility wins.

The numbers

• Up to 66% off (Compute SP, flexible) vs up to 72% (EC2 Instance SP, locked to family + region).
• How it applies: AWS meters compute hourly; usage up to your commit gets the discount, overage bills on-demand; under-use still pays the full commit.
• Payment: All Upfront (biggest discount), Partial Upfront, No Upfront (smallest, most popular for cash flow — still 50–60% off).
• Coverage target ~70–80% of compute spend; utilization target ~90%+ (Cost Explorer shows both).
• Field examples: a multi-region SaaS committed $200/hr globally and let coverage shift with a viral Asia launch, no exchanges; a MedTech team's single Compute SP saved 55% across EC2 and Lambda as it migrated batch jobs to serverless; a seasonal e-commerce shop committed $45/hr to cover 90% of its baseline and let December's $150/hr spike ride on-demand.

Do this

1. Rightsize first (with Compute Optimizer) — committing to oversized capacity locks in waste.
2. Commit to the baseline, not the peak — the steady $/hour you'll use even in slow periods; use Cost Explorer's Savings Plan recommendation (linked-account scope = all accounts in an Org).
3. Start conservative and 1-year if unsure — cover ~30–40% even for unpredictable workloads (still beats on-demand), buy more later; plans queue and stack automatically.
4. Pick payment by cash flow — No Upfront preserves liquidity at ~50–60% off; All Upfront for the deepest discount.
5. Check utilization monthly — consistently under 90% means you over-committed; exclude Spot usage (SPs don't cover it) from your commitment math.

Gotchas

• Under-use still bills the full commit — a $500/hr plan on $300/hr usage wastes $200/hr; start small, scale up.
• Doesn't cover Spot — leave Spot workloads out of the commitment calculation.
• Stacks with RIs, not exclusive — Savings Plans cover whatever existing Reserved Instances don't.
• Centralize purchases in an Org — one plan sized to Org-wide usage beats fragmented per-account commitments (see Consolidated Billing).

Skip this if

• You're 100% certain your instance family and region won't change for the term and want the maximum discount — EC2 Instance Savings Plans reach ~72%, and traditional Reserved Instances suit specialized/stable fleets.
• Your compute is almost entirely Spot or genuinely unpredictable with no stable baseline — there's little to commit. Stack this on top of Graviton moves for compounding savings, and compare against the broader Savings Plans overview.