Direct answer: cloud commitment discounts (AWS Savings Plans, Azure Reservations/Savings Plan, GCP CUDs) are not "the more you buy, the more you save"; they are "discount in exchange for coverage, with utilization as the safety net". The official ceilings are: AWS Savings Plans up to 72%, Azure Reserved VM up to 72% (Azure savings plan for compute up to 65%), and GCP CUDs with 1/3-year terms (actual rates depend on service/resource/region/term; this article does not quote an unverified single percentage). The decision rule is simple: if the discount is d, the break-even commitment utilization is about 1−d—a 20% discount needs at least 80% utilization, 30% needs 70%, 40% needs 60%, and 72% needs only 28%. So first calculate "stable baseline × coverage", then choose 1-year/3-year, upfront/partial/monthly, and finally monitor to avoid unused commitments.
This guide is for CFO/FinOps, cloud platform, procurement and architecture teams. It compares AWS Savings Plans, Azure Reservations/Savings Plan and GCP CUDs: mechanics, term/payment options, coverage and break-even, over-commitment risk, sharing scope and a 90-day adoption checklist. Rules come from the AWS Savings Plans page, Azure Reservations page and Google Cloud CUD documentation (all accessed 8 November 2026). Official mechanics facts, editorial calculations and items to confirm are kept separate; actual discounts follow each cloud console/contract quote.
(1) The break-even utilization for a discount d is about 1−d: 20% needs 80%, 30% needs 70%, 40% needs 60%, 72% needs only 28%. (2) Cover 60-80% of the stable baseline first, not peaks/new business; keep 20-40% on-demand/Spot. (3) Term/payment affect cash flow and flexibility: 1 year is safer, 3 years discounts more but locks longer; monthly payment usually costs no extra, but check currency/contract.
How the three work: Savings Plans / Reservations / CUDs
All three trade a usage/spend commitment for a discount, but flexibility differs: AWS Savings Plans have 1/3-year terms and payment options (all/partial/no upfront), and distinguish more flexible Compute Savings Plans from instance-family-bound EC2 Instance Savings Plans; Azure has Reserved VM Instances (region/instance/term) and a more flexible savings plan for compute; GCP has resource-based CUDs (resource-bound) and spend-based CUDs/Compute flexible commitments (spend commitment covering multiple services). Actual discounts are not public fixed values—the console/contract quote is authoritative; the official pages give "up to" ceilings.
| Item | AWS | Azure | GCP |
|---|---|---|---|
| Product | Savings Plans (Compute/EC2 Instance, etc.) | Reserved VM Instances; savings plan for compute | Resource-based CUDs; spend-based CUDs/flexible commitments |
| Term | 1 year / 3 years | 1 year / 3 years | 1 year / 3 years |
| Payment | All/partial/no upfront | Upfront or monthly (monthly usually no extra fee) | Monthly billing (spend/resource commitment) |
| Flexibility | Compute more flexible; EC2 Instance family-bound | RI region/instance-bound; savings plan more flexible | Resource-based bound; spend-based/flexible more flexible |
| Official ceiling | Up to 72% | Reserved VM up to 72%; savings plan up to 65% | By service/resource/term; console authoritative |
Sources: AWS Savings Plans page, Azure Reservations page, Google Cloud CUD documentation (accessed 8 November 2026). Editorial summary; "up to" is the official promotional ceiling; actual discounts depend on service/region/instance/term/payment/contract.
Break-even: how much utilization a discount d needs
A commitment discount means "you pay the commitment price whether you use it or not". If you commit 1 unit at discount d, the commitment cost is 1−d; if you actually use U, the on-demand cost would be U. To avoid losing money, U ≥ 1−d. So: 20% discount → 80% utilization; 30% → 70%; 40% → 60%; 50% → 50%; 72% → 28%. This rule matters more than "higher discount is better": if historical utilization is only 55%, a 40% discount (requiring 60%) can still lose money.
| Discount d | Break-even utilization 1−d | Note |
|---|---|---|
| 20% | 80% | Below 80% utilization, the commitment loses |
| 30% | 70% | Realistic range for many teams |
| 40% | 60% | Needs a high stable baseline |
| 50% | 50% | Break-even at half utilization |
| 72% | 28% | Ceiling discount; only stable workloads get it |
Editorial calculation: commitment cost = 1−d; on-demand cost = U; no loss when U ≥ 1−d. Excludes upfront cash flow, currency, tax, sharing scope and sunk cost of unused commitments.
Coverage scenarios: savings on $100k/month on-demand spend
Assume $100k/month on-demand spend, commitment coverage of 50%/70%/90%, and discounts of 20%/30%/40%; monthly savings = on-demand spend × coverage × discount. 70% coverage + 30% discount saves about $21k/month, $252k/year; 90% coverage + 40% discount saves about $36k/month, $432k/year. Higher coverage saves more but increases over-commitment risk; most teams start at 60-80% of the stable baseline, not 100%.
| Coverage | 20% discount | 30% discount | 40% discount |
|---|---|---|---|
| 50% | $10k/mo ($120k/yr) | $15k/mo ($180k/yr) | $20k/mo ($240k/yr) |
| 70% | $14k/mo ($168k/yr) | $21k/mo ($252k/yr) | $28k/mo ($336k/yr) |
| 90% | $18k/mo ($216k/yr) | $27k/mo ($324k/yr) | $36k/mo ($432k/yr) |
Editorial calculation: monthly on-demand spend = 100,000; savings = 100,000 × coverage × discount; excludes unused commitments, upfront cash flow, currency, tax, sharing scope and eligible-service limits. Discounts are scenario assumptions, not any vendor quote.
Over-commitment: how you lose when utilization is low
If you commit $80k/month at a 30% discount, the commitment cost is $56k/month; if you actually use $60k, on-demand would have been $60k, so the commitment still saves $4k; if you only use $50k, the commitment ($56k) is more expensive than $50k, so you lose $6k. That is the 70% break-even (30% discount). Over-commitment usually comes from committing to peaks, counting new business in the baseline, instance-family/region changes, ineligible services, not enabling organizational sharing, or currency/contract constraints.
| Commit $80k/mo at 30% | Actual use | Commitment cost | Vs on-demand |
|---|---|---|---|
| Commitment cost $56k | $80k | $56k | Save $24k |
| Commitment cost $56k | $60k | $56k | Save $4k |
| Commitment cost $56k | $50k | $56k | Lose $6k |
Editorial calculation: commitment cost = commitment × (1−discount); excludes upfront/monthly cash flow, sharing, currency and tax. Actual bills follow the vendor.
How to choose: term, payment, flexibility and sharing
Choose commitment by "stability", not "discount": a baseline stable for 12+ months fits 1 year; only 3-year visibility with stable instance families/regions justifies 3 years. Payment follows cash flow: all-upfront usually discounts most but ties up cash; partial/monthly is more flexible, and Azure monthly usually costs no extra. Flexibility follows change probability: Compute Savings Plans/Azure savings plan/GCP spend-based fit teams that change instance families/regions; bound RIs/EC2 Instance/resource-based CUDs discount more but lock more. Sharing scope matters: organization/billing-account sharing improves utilization but needs cost allocation/showback.
| Decision | Safer choice | Higher-discount choice |
|---|---|---|
| Term | 1 year | 3 years (high visibility) |
| Payment | Monthly/partial upfront | All upfront |
| Flexibility | Compute/savings plan/spend-based | EC2 Instance/RI/resource-based |
| Coverage | 60-80% of stable baseline | 90%+ (higher risk) |
| Sharing | Organization sharing + allocation | Single account (lower utilization) |
Editorial selection framework, not a vendor recommendation; products/discounts/sharing rules follow each cloud console and contract.
90-day adoption checklist
Use the first 90 days to close the "baseline, stability, coverage, term/payment, purchase, monitor" loop. Weeks 1-2 pull 12 months of on-demand bills/usage; weeks 2-4 identify the stable baseline (excluding peaks/new business/one-offs); weeks 4-6 calculate coverage/break-even; weeks 6-8 choose term/payment/flexibility/sharing; weeks 8-10 buy in batches; weeks 10-12 monitor utilization/savings/expiry. Do not buy 100% at once; start at 50-70%, verify utilization for 1-2 months, then add.
| Phase | Action | Output |
|---|---|---|
| Weeks 1-2 | Pull 12 months of on-demand bills/usage/services/regions/instance families | Cost baseline |
| Weeks 2-4 | Identify stable baseline; exclude peaks/new business/one-offs | Committable baseline |
| Weeks 4-6 | Calculate coverage, break-even and scenario savings | Coverage plan |
| Weeks 6-8 | Choose term/payment/flexibility/sharing; check cash flow | Procurement plan |
| Weeks 8-10 | Buy in batches (50-70% first); record expiry/renewal | First commitments |
| Weeks 10-12 | Monitor utilization/savings/unused; quarterly review | Dashboard, review |
Editorial adoption advice, not an official template; purchase/term/payment/sharing follow enterprise finance and vendor contracts.
FAQ
How much can commitment discounts save?
AWS Savings Plans up to 72%, Azure Reserved VM up to 72% (savings plan for compute up to 65%); GCP CUDs depend on service/resource/term, with the console authoritative. Actual discounts depend on service, region, instance family, term, payment and contract.
What coverage is appropriate?
Most teams start at 60-80% of the stable baseline and keep 20-40% on-demand/Spot; do not commit peaks or new business. Start at 50-70% and add after verifying utilization.
How do we calculate break-even?
Break-even utilization for discount d is about 1−d: 20% needs 80%, 30% needs 70%, 40% needs 60%, 72% needs 28%. Below that, the commitment may cost more than on-demand.
1 year or 3 years?
Look at visibility: stable for 12+ months, start with 1 year; only 3-year visibility and stable instance families/regions justify 3 years. 3 years discounts more but locks longer.
All upfront or monthly?
All upfront usually discounts more but ties up cash; monthly/partial is more flexible. Azure monthly usually costs no extra; AWS has all/partial/no upfront; GCP bills monthly. Choose by cash flow and currency.
Can commitments be shared?
AWS/Azure/GCP all support organization/billing-account sharing (rules differ). Sharing improves utilization but needs cost allocation/showback so departments recognize their usage.
What happens to unused commitments?
Unused commitment is still billed (the commitment is pay-whether-used-or-not). If utilization is below 1−d, the commitment can be more expensive than on-demand; calculate the baseline, buy in batches and monitor utilization.
What costs are not covered?
Usually data transfer, support plans, taxes and some licenses/marketplace purchases are not covered; check the vendor page/console. Do not treat the entire bill as a committable base.
Research method and limitations
This article separates "official mechanics facts" from "editorial calculations." AWS's 72% ceiling, 1/3-year terms, payment options and Savings Plans types come from the AWS Savings Plans page. Azure Reserved VM's 72%, savings plan's 65%, 1/3-year terms and monthly payment at no extra cost come from the Azure Reservations page. GCP CUD's 1/3-year terms and resource-based/spend-based/flexible commitments come from the Google Cloud CUD documentation (accessed 8 November 2026). The break-even, coverage scenarios, over-commitment example and 90-day checklist are editorial; discounts are scenario assumptions, not any vendor quote.
Eight limitations: (1) this is not financial/procurement advice; (2) no specific SKU/region actual discount is quoted; (3) upfront cash flow, currency, tax, sharing and contract terms are excluded; (4) not all commitment types (database/storage/AI-specific) are covered; (5) no real bill/utilization model is included; (6) the accounting impact of sharing/allocation is not assessed; (7) transfer/exchange/cancellation rules are not covered; (8) prices/products change, so check the console/contract before purchase.
Sources and basis
- AWS, Savings Plans: aws.amazon.com/savingsplans. Up to 72%, 1/3-year terms, payment options, plan types. Accessed 8 November 2026.
- Microsoft Azure, Reservations: azure.microsoft.com/en-us/pricing/reservations. Reserved VM up to 72%, savings plan up to 65%, 1/3-year terms, monthly payment. Accessed 8 November 2026.
- Google Cloud, Committed use discounts: cloud.google.com/docs/cuds. 1/3-year terms, resource-based/spend-based/flexible commitments. Accessed 8 November 2026.
Data cut-off 8 November 2026; discounts/terms follow each cloud console, contract and official page; this is not financial, procurement or legal advice.