The difference between a 30% savings and a 60% savings often comes down to one decision: how much of your cloud workload should you commit to upfront? This guide covers the commitment dilemma, sizing frameworks, multi-cloud portfolio strategies, and 8 tactics to maximize discounts while minimizing waste.
Every enterprise using cloud faces the same trade-off: pay premium on-demand rates for flexibility, or lock in discounted rates by committing capacity upfront. The problem is real—most companies either over-commit (purchasing unused capacity they waste money on) or under-commit (paying inflated on-demand premiums for variable loads).
On-demand pricing typically runs 2–4x higher than committed rates across all three hyperscalers. For example, a single AWS m5.2xlarge instance costs approximately $0.384/hour on-demand but just $0.114/hour with a 3-year Savings Plan—a 70% discount. An Azure Standard_D4s_v5 is $0.376/hour on-demand but $0.118/hour with a 3-year commitment. That gap is your savings opportunity, but also your risk zone: over-commit and you're locked into paying for capacity you don't use; under-commit and you're leaving 30–60% savings on the table.
The challenge is that most organizations lack visibility into their true stable workload baseline. They look at current spend, don't know what portion is temporary, and end up guessing. This guide provides a data-driven framework to eliminate the guesswork.
Each hyperscaler offers different commitment instruments with varying flexibility and discount structures. Understanding what each offers is the first step to building the right portfolio.
Want independent help negotiating better terms? We rank the top advisory firms across 14 vendor categories — free matching, no commitment.
| Provider | Instrument | Discount vs OD | Flexibility | Min Term | Upfront |
|---|---|---|---|---|---|
| AWS | Reserved Instance (Std) | 30–40% | None (rigid) | 1 or 3yr | Full/Partial |
| AWS | Reserved Instance (Conv) | 25–35% | Family only | 1 or 3yr | Full/Partial |
| AWS | Savings Plan | 24–30% | High (family/region) | 1 or 3yr | Partial |
| AWS | EDP | 5–20% | Full (no commitment) | 1yr | None |
| Azure | Reserved Instance | 30–40% | Series only | 1 or 3yr | Full/Partial |
| Azure | Savings Plan | 20–28% | High (family/region) | 1 or 3yr | Partial |
| Azure | MACC | 10–20% | Full (services vary) | Custom | Credit basis |
| GCP | Committed Use Discount | 28–55% | None (rigid) | 1 or 3yr | Full |
| GCP | Sustained Use Discount | 20–30% | Full (auto) | None | None |
| GCP | Commit Program | 10–15% | High (services) | Custom | None |
The single biggest mistake enterprises make is treating commitments as an all-or-nothing binary. In reality, the optimal strategy is a tiered portfolio: a committed baseline layer for stable, predictable workloads, plus a flexible layer for variable demand.
Get the IT Negotiation Playbook — free
Used by 4,200+ IT directors and procurement leads. Oracle, Microsoft, SAP, Cloud — all covered.
Commit to 70% of your stable baseline load; leave the remaining 30% on flexible pricing (on-demand, Spot, or variable rates). This approach balances two competing risks:
Empirically, 70% is the "Goldilocks zone" for most enterprises. It captures the majority of savings while protecting against business volatility (seasonal peaks, mergers, new projects, cloud migrations).
Your commitment portfolio should reflect your actual workload distribution:
A $2M/year cloud customer might structure: $1.2M baseline (commit 70% = $840K), $600K variable (on-demand), $200K tactical (short-term). This allows them to capture ~$200–250K in annual savings (30% discount on committed $840K) while preserving flexibility for the remaining $800K of workload.
The most reliable way to size commitments is the percentile methodology using 30, 60, and 90-day lookback windows. This removes anomalies (spikes, one-off projects) and identifies your true stable load.
Pull the last 90 days of daily costs (or hourly if more granular) from AWS Cost Explorer, Azure Cost Management, or GCP Cost Management. Exclude non-recurring costs (data transfer, enterprise licenses, one-time migrations).
Sort your 90 days of daily costs from lowest to highest. The 75th percentile is the cost level where 75% of your days fall below and 25% above. This is your safe baseline commit point.
Why the 75th percentile? The p50 (median) often misses seasonal variation; the p90 is too conservative and wastes savings. The p75 is empirically optimal: it captures most stable load while excluding spikes.
For a more granular approach, calculate the 75th percentile separately for:
This allows you to commit only to the most predictable services and leave variable services flexible.
AWS Cost Explorer has a built-in "RI Purchase Recommendations" report that suggests RI quantities based on your usage. However, take AWS recommendations with caution—they typically suggest 40–50% higher commitments than your safe baseline, because AWS benefits from over-purchases (unused commitments are revenue they keep).
Our recommendation: use AWS recommendations as a starting point, but reduce the suggested commitment by 20–30% to account for business volatility.
Azure Advisor analyzes your usage and recommends RI quantities; GCP Recommender does the same for CUDs. Both have similar bias towards over-commitment. Apply the same 20–30% reduction to be conservative.
| Metric | p50 Baseline | p75 Baseline | p90 Baseline | Recommended Commit |
|---|---|---|---|---|
| Avg Daily Cost | $4,200 | $4,600 | $5,100 | $3,220 (70% of p75) |
| Monthly Cost | $126,000 | $138,000 | $153,000 | $96,600 |
| Annual Cost (OD) | $1,512,000 | $1,656,000 | $1,836,000 | $1,159,200 |
| 3-Yr SP Discount (26%) | $393,120 | $430,560 | $477,360 | $301,392 |
| Annual Savings | $131,040 | $143,520 | $159,120 | $100,464 |
| Remaining OD | $1,119,000 | $1,112,400 | $1,122,600 | $857,808 (30% flex) |
In this example, committing at the p75 baseline (70% = $96.6K/month) saves $100K+/year while preserving flexibility for the remaining 30% of workload variability.
AWS customers often ask: should we buy Savings Plans or Reserved Instances? The answer depends on your workload stability and need for flexibility.
Savings Plans are the modern commitment vehicle on AWS. They commit you to a dollar amount of compute spend (not specific instances), giving you flexibility across instance families, regions, sizes, and OS. For example, your $1M Savings Plan can shift between m5.2xlarge in us-east-1, c6i.4xlarge in eu-west-1, or any other compute combination.
Discount: 24–30% for 3-year plans (slightly lower than RIs, but flexibility is worth it).
When to buy: If you have diverse workloads, plan infrastructure changes, or migrate between regions/instance families.
Standard RIs lock you into a specific instance type, size, and region (or AZ). They deliver 30–40% discounts, about 3–5% better than Savings Plans, but offer zero flexibility.
When to buy: Only if you have a stable, single-region, single-instance-type workload that will run unchanged for 3 years (rare).
Convertible RIs allow you to change instance family during the commitment term (e.g., switch from m5 to m6 if newer generations launch). They cost 25–35%—less than Savings Plans but more flexible than Standard RIs.
When to buy: If you want instance family flexibility but expect no regional changes.
Buy 80–90% Savings Plans (for maximum flexibility) and 10–20% Convertible RIs (if you have legacy workloads requiring family upgrades). Avoid Standard RIs unless you have extreme confidence in workload stability.
AWS Marketplace allows resale of RIs, but resold RIs have no flexibility (locked to original instance family/region/AZ). Avoid buying RIs on the marketplace unless you're liquidating unused capacity at deep discounts.
Azure's commitment strategy is more complex because RIs, Savings Plans, and Hybrid Benefit (BYOL) interact in layered ways.
Commit to compute capacity (vCPU count + memory within a series) for 1 or 3 years. You choose:
Commit to hourly compute spend across any VM family, region, or OS (Windows vs. Linux). Think of it as Azure's answer to AWS Savings Plans.
If you own Windows Server or SQL Server licenses on-premises (via Software Assurance), you can reuse them in Azure at no extra cost. This stacks on top of RIs for a compounding benefit.
Example: Windows Server Standard license costs $608/2-year; reuse it in Azure saves ~$30–50/month on licensing. On top of a 35% RI discount, your total savings is ~50%.
Key point: AHB requires active Software Assurance; without it, you're just paying Windows licensing on top of compute. Confirm your license status before architecting around AHB.
For large customers, Azure offers MACC tiers ($50K, $100K, $500K+/year commitments) that apply credits across compute, storage, databases, and some third-party services. MACC discounts typically 10–20% off list price.
MACC advantage: Service flexibility—you can shift credits between services as needs change. MACC disadvantage: credits expire if unused by end-of-year, so purchasing requires visibility into full-year spend.
Unlike AWS RIs (no refunds), Azure allows exchanging RIs for larger/smaller sizes within the same family and canceling RIs for refunds (minus a 12% penalty on 3-year plans). This makes Azure RIs less risky, but the penalty still makes accurate sizing important.
GCP's commitment landscape is different because of SUDs (Sustained Use Discounts), which apply automatically with zero commitment.
You commit to a specific machine type (e.g., n1-standard-4), CPU count, memory, or resource for 1 or 3 years. Discounts range 28–55% depending on resource type and term.
Trade-off: High discount but rigid—if you over-commit or your needs change, you're locked in with no refund mechanism (unlike Azure).
Automatic discounts applied to any resource you use >25% of the month: 20–30% discount, no commitment, no upfront cost, applies automatically. The longer you run a machine, the higher the discount tier.
Example: Running a machine the entire month gets a 30% SUD; running it 50% of the month gets 15% SUD.
Use SUDs first: Free discount, automatically applied—no reason not to. If your machine runs >25% of the month, you get SUDs with zero effort.
Use CUDs for stable workloads: On top of SUDs, if you're confident about a specific machine type for 3 years, buy CUDs for the extra 10–20% discount. A machine with 30% SUD + 28% CUD gets ~50% total discount.
Avoid CUDs for uncertain workloads: If you're still optimizing or expect infrastructure changes, skip CUDs and ride the automatic SUD savings until you're confident.
For customers spending >$12,500/year, GCP offers tiered spending commitments ($12.5K, $25K, $50K+) with 10–15% negotiated discounts. This is an alternative to per-resource CUDs if you want flexibility across services.
Beyond on-demand commitments, all three hyperscalers offer enterprise-wide discount programs negotiated with sales. These are often overlooked but can save 5–20% across your entire cloud bill.
Eligibility: Minimum $1M/year committed spending.
Discount structure: Tiered discounts on all services (not just compute), typically 5–10% at $1M tier, 10–15% at $5M tier, 15–20% at $10M+ tier. Discounts apply to all consumption (RIs, Savings Plans, on-demand).
Key advantage: Applies to non-compute services (RDS, ECS, Lambda, etc.) where you can't use RIs/SPs.
When to negotiate: As soon as you hit $1M annual spend. Most enterprises don't ask, so AWS rarely volunteers.
Eligibility: Minimum $50K/year (though $100K+ more common).
Structure: You purchase "commitment credits" that apply to Azure consumption. MACC credits cover compute, storage, databases, and select third-party services.
Example: $500K MACC purchase gives you $500K in credits. If you spend $600K in a year, you use $500K in credits and pay $100K on-demand.
Discount reality: MACC commits don't directly translate to a % discount, but empirically provide 10–20% savings vs. pure on-demand. The savings come from removing overage on-demand rates, which are priced high.
Risk: Unused credits expire at year-end (typically Dec 31), so you must forecast annual spend accurately. Over-committing MACC is waste; under-committing leaves discounts uncaptured.
Eligibility: Minimum $12.5K/year; large enterprise tiers at $25K, $50K, $100K+.
Discount: 10–15% on most services (not just compute). Smaller discount than AWS EDP, but GCP commit is less common, so there may be negotiation room.
Negotiate EDP/MACC deals before fiscal year-end when vendors are pressured to close discounts. October–November are peak negotiation windows. Avoid Q1 negotiations; vendors are more conservative early in the fiscal year.
The timing of commitment purchases can have a 5–10% impact on your total savings. Hyperscaler sales organizations have quarterly/annual targets, and they're more flexible (willing to negotiate deeper discounts) at specific times.
The biggest timing mistake is waiting until your RI/Savings Plan expires and scrambling to renew. Instead, purchase the new commitment 30–90 days before expiry. Why?
If your company's fiscal year differs from the calendar year, align your cloud commitments to match. Why? Because your CFO budgets cloud spend by fiscal year, and commitment terms that straddle fiscal years create accounting complexity and budget planning friction.
Don't wait until Q4 (Oct–Dec) to commit if you haven't yet. Q4 is when vendors are flooded with renewal requests and have less flexibility. Similarly, vendors who haven't hit their annual quotas by Q3 are more aggressive on pricing, so October is often better for negotiations than November/December.
AWS re:Invent (typically November) is when AWS announces new instance types, services, and pricing. If you're on a 3-year RI, don't buy RIs immediately before re:Invent (you'll be locked into older instance families). Wait until post-re:Invent to see the new landscape, then commit.
Many large enterprises run multi-cloud (AWS primary, Azure secondary, GCP for analytics). The challenge: committing to all three at maximum levels locks you into $3–5M/year in unused capacity if business priorities shift.
Instead of committing equally to all three clouds, allocate commitments by strategic importance:
AWS Savings Plans can shift between services (compute, database, etc.) but not between clouds. Azure RIs are subscription-specific. GCP CUDs are machine-specific. The bottom line: cloud commitments are not portable between hyperscalers. If you migrate a workload from AWS to Azure, your AWS Savings Plans don't follow.
Plan commitments assuming workloads stay in-cloud for the commitment term. If you're actively migrating between clouds, reduce commitment terms to 1-year instead of 3-year to minimize lock-in risk.
Your commitment strategy isn't "set and forget." The cloud is dynamic—new services launch, workloads shift, teams migrate systems. Every quarter, audit your commitments and rebalance.
Every 90 days, run these reports:
If your quarterly audit reveals misalignment:
If you've over-committed and have 6+ months of unused RI capacity, post them to the AWS Marketplace. Buyers get a discount (e.g., you paid $10K for a 3-year RI, sell it for $6K with 1.5 years remaining). You recover 60% of sunk cost; better than watching it burn.
Need help optimizing your cloud commitment strategy?
Our cloud negotiation experts help enterprises commit strategically, size accurately, and save 30–60% on cloud infrastructure.
Your commitment strategy is the single biggest lever in cloud cost optimization. Get it right and you save 30–60% annually. Get it wrong and you waste millions on unused capacity or over-priced on-demand consumption.