Updated 14 Aug 2026 • 4 mins read
Reserved Instances vs Savings Plans vs Spot: The 2026 Guide
Reserved Instances, Savings Plans, and Spot are the three ways to pay less than on-demand for AWS compute. RIs and Savings Plans trade a commitment for up to 72 percent off; Spot trades reliability for up to 90 percent. This guide explains how each works and when to use which.
On-demand pricing is the rate you pay for the privilege of not thinking about it. The moment you are willing to make a promise about your usage, or to give AWS the right to take capacity back, the price drops, sometimes by a little, sometimes by 90 percent. Reserved Instances, Savings Plans, and Spot are the three ways to make that trade, and choosing between them is one of the highest-leverage cost decisions a team makes, because the same workload can cost four times as much on the wrong one. It is the practical core of cloud pricing models.
This guide compares Reserved Instances, Savings Plans, and Spot in 2026: what each actually is, how much it saves, what you give up, and the simple rule for deciding which covers which part of your workload.
The short version Reserved Instances commit you to a specific instance configuration for 1 or 3 years for up to 72 percent off. Savings Plans commit you to an hourly dollar amount, not a specific instance, for similar savings (Compute Savings Plans up to 66 percent, EC2 Instance Savings Plans up to 72 percent) with far more flexibility. Spot sells spare capacity for up to 90 percent off but can reclaim it with two minutes' notice. Most teams layer all three: Savings Plans for the steady baseline, Spot for interruptible work, RIs for specific reserved needs.
Reserved Instances (RIs)
Reserved Instances are the original commitment discount. You agree to a specific configuration, instance family, size, region, and sometimes operating system and tenancy, for a 1-year or 3-year term, and in return you pay up to 72 percent less than on-demand. Standard RIs give the deepest discount but lock you to that configuration; Convertible RIs give a smaller discount (around 54 percent) but let you exchange for a different configuration later. Payment can be all upfront (best discount), partial upfront, or no upfront.
RIs still matter in 2026, but for pure EC2 compute, Savings Plans have largely superseded them because they offer similar savings with less lock-in. Where RIs remain the tool of choice is for services Savings Plans do not cover, such as RDS, ElastiCache, Redshift, and OpenSearch reserved nodes, and for teams that want capacity reservations tied to the discount.
Savings Plans
Savings Plans replace the rigid, instance-specific commitment with a flexible, dollar-based one: you commit to spending a certain amount per hour (for example $10/hour) for 1 or 3 years, and any usage up to that amount is discounted automatically. There are two kinds. Compute Savings Plans are the most flexible, applying across instance families, sizes, regions, operating systems, tenancy, and even Fargate and Lambda, for up to 66 percent off. EC2 Instance Savings Plans lock you to an instance family in a region for a deeper discount, up to 72 percent, while still letting you move sizes within that family.
The flexibility is the point: because a Compute Savings Plan follows your usage as you re-architect, migrate regions, or switch to Graviton, it keeps saving without renegotiation. For managing commitments without over-buying, see our discount manager guide.
Spot Instances
Spot sells AWS's spare capacity at up to 90 percent off on-demand, the steepest discount available, with one catch: AWS can reclaim the instance with a two-minute warning when it needs the capacity back. That makes Spot unsuitable for a stateful database, but ideal for anything interruptible and fault-tolerant, batch processing, CI/CD, big-data jobs, containerized and stateless workloads, and machine-learning training that checkpoints. Modern tooling, Spot Fleets, capacity-optimized allocation, and Kubernetes autoscalers, spreads workloads across many instance pools to minimize interruptions, making Spot far more practical than its reputation suggests.
Head-to-Head
The three are not competitors so much as tools for different parts of the same bill.
| Model | Savings | Commitment | Flexibility | Best for |
|---|---|---|---|---|
| Reserved Instances | Up to 72% | 1 or 3 years; specific configuration | Low to medium | RDS, ElastiCache, Redshift, and other reservation-eligible workloads |
| Compute Savings Plans | Up to 66% | 1 or 3 years; committed $/hour | High — applies across instance families, regions, Fargate, and Lambda | Steady but evolving compute baselines |
| EC2 Instance Savings Plans | Up to 72% | 1 or 3 years; committed $/hour within a family | Medium — sizes can change within the family | Stable workloads in a known instance family |
| Spot | Up to 90% | No long-term commitment; capacity can be reclaimed | Full, but workloads must tolerate interruption | Batch, CI/CD, stateless workloads, and ML training |
How to Combine Them
The teams that get the most from these do not pick one; they layer all three against a simple rule: cover the floor, flex the middle, gamble the top. Buy Savings Plans (usually Compute) to cover the steady baseline of usage you are confident will persist, typically 60 to 80 percent of always-on compute, so you never over-commit. Run all interruptible and fault-tolerant work on Spot to capture the deepest discount. Use Reserved Instances for the services Savings Plans do not cover and for any capacity you must guarantee. Leave the unpredictable remainder on on-demand. Start conservative on commitments, watch your coverage and utilization, the FinOps KPIs that keep this honest, and increase the commitment as confidence grows.
Choosing Your Mix
There is no single best AWS pricing model, only the right model for each slice of your workload. Think of your compute as a stack: a stable base that never turns off, a flexible middle that shifts as you build, and a spiky, interruptible top. Savings Plans are built for the base, Spot for the interruptible top, and Reserved Instances for the reserved and non-EC2 corners the others miss. Match each layer to its model, forecast the baseline you are committing to with our cloud cost forecasting guide, and back it with the broader practice in AWS cost optimization with FinOps. Do that and the same infrastructure can cost a fraction of the on-demand sticker, not because you negotiated harder, but because you stopped paying on-demand rates for usage that was never on-demand in nature.
FAQs
Reserved Instances commit you to a specific instance configuration for a deeper, exact-match discount. Savings Plans commit you to an hourly dollar amount that applies automatically across families, regions, and even Fargate and Lambda, trading a little discount for much more flexibility.
Reserved Instances save up to 72 percent, Compute Savings Plans up to 66 percent, EC2 Instance Savings Plans up to 72 percent, and Spot up to 90 percent. The deeper the commitment or the more interruption you accept, the larger the discount.
For interruptible, fault-tolerant workloads: batch processing, CI/CD, big-data jobs, stateless and containerized services, and machine-learning training that checkpoints. Avoid Spot for stateful databases or anything that cannot tolerate a two-minute reclaim warning.
For pure EC2 compute, Savings Plans have largely replaced them by offering similar savings with less lock-in. Reserved Instances remain the right tool for RDS, ElastiCache, Redshift, and OpenSearch reserved nodes, and when you need a capacity reservation tied to the discount.