Reserved Instances vs. On-Demand: What PE Firms Need to Know
Commitment-based pricing is one of the most impactful levers in cloud cost optimization, yet it remains one of the least understood among PE professionals. This is not surprising -- the terminology is technical, the options are numerous, and AWS does not make it particularly intuitive. But the financial impact is too significant to ignore: proper commitment strategy can reduce compute costs by 30-60%, often representing the single largest savings opportunity in a portfolio company's cloud environment.
This guide explains commitment-based pricing in financial terms that PE investors and operating partners can use to evaluate targets, model savings, and oversee implementation.
The Fundamental Trade-Off
AWS cloud pricing starts with On-Demand: you pay an hourly rate for compute capacity with no commitment. You can start and stop instances at any time, scale up or down, and pay only for what you use. On-Demand is the list price -- the starting point from which all discounts are measured.
Commitment-based pricing offers a discount in exchange for a commitment to a certain level of usage over a one-year or three-year term. The analogy is straightforward: it is the difference between paying month-to-month rent and signing a lease. The landlord gives you a better rate in exchange for guaranteed occupancy.
The core trade-off: Lower per-unit cost in exchange for reduced flexibility and the risk that your usage falls below your commitment level.
Here are the three main commitment vehicles, in order from most flexible to most restrictive:
Compute Savings Plans
Introduced in 2019, Compute Savings Plans are the most flexible commitment option and the one I recommend as the starting point for most companies.
How they work: You commit to a consistent amount of compute spend per hour (measured in dollars, not specific resources) for a 1-year or 3-year term. In exchange, you receive a discount on all compute usage -- including EC2, Fargate, and Lambda -- regardless of instance family, size, operating system, or region.
Example: You commit to $50/hour in compute spend for 1 year. Your On-Demand compute runs approximately $80/hour. Your effective rate drops to approximately $50/hour for the committed portion plus On-Demand rates for the remaining $30/hour. Net savings: approximately 25-30% on total compute.
Discount levels (approximate):
- 1-year, no upfront payment: 20-25% discount
- 1-year, all upfront payment: 28-33% discount
- 3-year, no upfront payment: 35-40% discount
- 3-year, all upfront payment: 50-55% discount
Key advantage: Maximum flexibility. If you change instance types, move to containers, adopt serverless, or shift regions, the Savings Plan still applies. This makes Compute Savings Plans ideal for companies undergoing architectural changes or rapid growth.
Key risk: If your compute usage drops below the committed level, you are paying for unused capacity. However, since the commitment is measured in dollars across all compute (not tied to specific instances), this risk is relatively low for any company with stable or growing compute needs.
EC2 Instance Savings Plans
EC2 Instance Savings Plans offer deeper discounts than Compute Savings Plans but are tied to a specific instance family in a specific region.
How they work: You commit to usage of a specific instance family (e.g., c6g in us-east-1) for a 1-year or 3-year term. In exchange, you receive a larger discount than Compute Savings Plans. The commitment applies regardless of instance size, OS, or tenancy within that family.
Discount levels (approximate):
- 1-year, no upfront: 28-33% discount
- 1-year, all upfront: 35-40% discount
- 3-year, no upfront: 45-50% discount
- 3-year, all upfront: 55-62% discount
Key advantage: Deeper discounts for workloads where you have high confidence in the instance family and region over the commitment term.
Key risk: If you migrate to a different instance family or region, the savings plan does not follow. This is a real risk for companies planning architectural modernization -- for example, migrating from c5 (Intel) to c7g (Graviton/ARM).
Reserved Instances (Legacy)
Reserved Instances (RIs) are the original commitment-based pricing model, predating Savings Plans. They are still available and still used, particularly for services where Savings Plans do not apply (RDS, ElastiCache, OpenSearch, Redshift).
How they work: You purchase a reservation for a specific instance type, in a specific region, for a 1-year or 3-year term. Standard RIs are locked to the exact configuration. Convertible RIs allow you to exchange for a different configuration of equal or greater value.
Key use cases: RIs remain the primary commitment vehicle for managed database services. An RDS Reserved Instance for a db.r6g.2xlarge can save 35-55% compared to On-Demand, depending on term and payment option.
Key limitation: Less flexible than Savings Plans for EC2. I generally recommend Savings Plans for compute and RIs only for database and cache services where Savings Plans do not apply.
Enterprise Discount Programs (EDPs)
For companies with significant aggregate AWS spend -- typically $1M+ annually -- AWS offers Enterprise Discount Programs. These are custom commercial agreements negotiated directly with the AWS sales team.
How they work: The company commits to a minimum annual spend level (measured in total AWS consumption, not specific services) for a 1-3 year term. In exchange, AWS provides a percentage discount on all services, applied on top of any other discounts (including Savings Plans and RIs).
Typical discount levels: 5-15%, depending on commitment size, term, and growth trajectory. A company committing to $3M annually for 3 years might negotiate an 8-12% EDP discount.
Key advantage: The discount applies to all AWS services, including those not covered by Savings Plans or RIs (S3, data transfer, Lambda, etc.). For companies with diverse AWS usage, the aggregate savings from an EDP can be substantial.
Key risk: If actual spend falls below the committed minimum, the company still owes the committed amount. This is a real financial liability that should be carefully modeled during diligence. I have seen companies sign EDPs based on aggressive growth projections that did not materialize, resulting in an overpayment obligation.
Evaluating a Target's Commitment Strategy During Diligence
When assessing a target company's cloud commitment strategy, here are the key metrics and questions:
Coverage ratio: What percentage of stable compute spend is covered by Savings Plans or RIs? For a well-managed environment, the target should be 60-75%. Below 40% indicates significant untapped savings. Above 85% risks over-commitment.
Commitment term mix: What is the balance between 1-year and 3-year commitments? Three-year terms offer deeper discounts but carry more risk. I generally recommend a 60/40 or 70/30 split favoring 1-year terms for most PE-backed companies, given the typical hold period and potential for operational changes.
Utilization rate: Are existing commitments being fully utilized? AWS provides a reservation utilization report showing what percentage of committed capacity is actually being used. Utilization below 80% indicates over-commitment -- the company is paying for more than it uses.
Expiration schedule: When do existing commitments expire? A wave of expirations can cause a sudden cost increase if not proactively managed. Conversely, commitments that extend well beyond your expected hold period may constrain a future buyer.
EDP terms: If an EDP is in place, what are the minimum spend commitments by year? Is the company tracking to meet them? What happens if spend falls below the minimum? What is the exit clause?
Financial Modeling for PE
When building cloud cost models for a PE investment, here is how to incorporate commitment-based pricing:
Step 1: Establish the On-Demand baseline. Determine what the company would be paying at full On-Demand rates. This is your starting point.
Step 2: Apply current commitment discounts. Factor in existing Savings Plans, RIs, and EDP discounts based on their terms and expiration dates.
Step 3: Model the optimization opportunity. If the target has low commitment coverage, model the savings from implementing appropriate commitments. Be conservative: assume 60% coverage with 1-year Compute Savings Plans as a starting point. This typically yields 15-20% savings on total compute spend.
Step 4: Account for commitment liabilities. Include any existing commitment obligations in your model. Three-year RIs and EDP minimums are real financial commitments that persist regardless of usage changes.
Step 5: Align commitment terms with your hold period. If your expected hold period is 3-5 years, be thoughtful about 3-year commitments made in year 3 or later. A buyer may not want to inherit commitment obligations that extend beyond their own investment horizon.
Common Mistakes to Avoid
Over-commitment: The enthusiasm to capture savings can lead to over-purchasing commitments. If usage drops -- due to optimization, customer churn, or architectural changes -- over-committed capacity becomes a wasted expense. Always maintain a 25-35% On-Demand buffer for flexibility.
Ignoring Savings Plans in favor of RIs: Some companies (and some consultants) still default to Reserved Instances for EC2. For most use cases, Compute Savings Plans or EC2 Instance Savings Plans offer comparable discounts with significantly more flexibility. Use RIs only for services where Savings Plans do not apply.
Neglecting the EDP opportunity: Companies spending $1M+ annually on AWS often do not realize they qualify for an EDP. This leaves 5-15% savings on the table across their entire AWS bill. During diligence, always check whether the target has an EDP and, if not, whether they qualify for one.
Not modeling the interaction between commitment types: Savings Plans, RIs, and EDPs layer on top of each other. The effective discount is not simply additive -- understanding the waterfall of how discounts apply requires careful analysis. Get this wrong and your savings projections will be inaccurate.
The Bottom Line
Commitment-based pricing is not glamorous. It does not require cutting-edge technology or complex architectural changes. But it is one of the most reliable, lowest-risk levers for reducing cloud costs in PE-backed companies.
For a company spending $2M annually on compute with no commitment coverage, implementing a thoughtful Savings Plan strategy can deliver $400K-$700K in annual savings. That is real EBITDA improvement with minimal implementation risk and a 2-3 week timeline from decision to execution.
It is, quite simply, the easiest money in cloud optimization. The only question is whether your portfolio companies are capturing it.
Ready to evaluate cloud economics in your next deal? Book a free discovery call to discuss your specific situation.