Reserved Instances vs. Savings Plans vs. Spot: The Complete AWS Discount Guide for 2026
Every AWS cost optimization conversation eventually collides with the same three-way decision: Reserved Instances, Savings Plans, or Spot. Get it right and you can shave 40-70% off On-Demand pricing. Get it wrong — over-commit to the wrong instance family, or run stateful workloads on Spot — and you end up with either wasted commitment or 2am pages from interrupted instances. Here's the decision framework we actually use with clients.
On-Demand: The Expensive Default
On-Demand pricing is the ceiling, not the target. If more than 20% of your steady-state compute is running On-Demand, you're almost certainly overpaying. We start every FinOps engagement by pulling 60-90 days of Cost and Usage Report (CUR) data to establish what's actually "steady state" before recommending any commitment — committing based on a single spiky month is how teams end up locked into capacity they don't need.
Reserved Instances: Still Relevant, Narrower Use Case
Reserved Instances (RIs) used to be the default cost lever. In 2026, we reserve RIs for a narrower set of cases:
- RDS and ElastiCache: Database Reserved Instances frequently still beat Savings Plans coverage for stable, predictable database workloads.
- Regional/AZ-specific capacity guarantees: If you need guaranteed capacity in a specific Availability Zone during high-demand periods, Zonal RIs provide that; Savings Plans do not guarantee capacity.
For general EC2 compute, Savings Plans have mostly superseded RIs because of their flexibility across instance families and regions.
Savings Plans: The Default for EC2, Fargate, and Lambda
Compute Savings Plans apply automatically across instance family, size, OS, and region — you commit to a $/hour spend target, not a specific instance type. This flexibility is why we default here for most compute:
- 1-year vs 3-year: We generally recommend starting with 1-year commitments for teams newer to commitment planning, moving to 3-year for workloads with proven multi-year stability (typically saves an additional 15-20% over 1-year terms).
- Coverage target: Aim to cover 70-80% of steady-state usage with commitments, leaving headroom for organic growth and seasonal spikes to run On-Demand or Spot.
Spot Instances: Free Money for Fault-Tolerant Workloads
Spot Instances offer discounts of 60-90% off On-Demand, in exchange for the possibility of a 2-minute interruption notice. The right candidates:
- CI/CD build runners
- Batch and ETL processing jobs
- Stateless web tier behind an Auto Scaling Group with mixed instance policies
- Non-critical dev/test environments
In one recent migration engagement, we moved roughly 50% of a client's stateless compute to Spot using EC2 Fleet with diversified instance pools (to reduce simultaneous interruption risk), which alone cut their compute bill by approximately 30% without any application changes.
A Practical Allocation Model
| Workload Type | Recommended Purchase Model |
|---|---|
| Production databases (RDS/ElastiCache) | Reserved Instances |
| Steady-state EC2/Fargate/Lambda | Compute Savings Plans |
| CI/CD, batch, stateless workers | Spot Instances |
| Unpredictable/new workloads | On-Demand (until 60-90 days of usage data exist) |
Don't Set and Forget
Commitment planning isn't a one-time exercise. Usage patterns shift as products evolve, and an RI or Savings Plan purchased for last year's traffic can quietly become under-utilized. We use FinOps Co-Pilot to continuously track commitment utilization for clients so under-covered or over-committed capacity gets flagged automatically rather than discovered at renewal time. If you want a second opinion on your current commitment mix, our team can run a free audit — reach out here.
Frequently Asked Questions
Should I buy Reserved Instances or Savings Plans in 2026?
For most workloads, Compute Savings Plans have replaced Reserved Instances as the default because they apply across instance families, regions, and even Fargate and Lambda. Standard RIs still make sense for stable RDS and ElastiCache workloads.
What percentage of workloads can safely run on Spot Instances?
We typically move 30-50% of stateless compute — CI/CD runners, batch processing, background jobs — to Spot. Stateful or interruption-sensitive workloads should stay on On-Demand or committed capacity.
How does FinOps Co-Pilot help with commitment planning?
It analyzes 30-90 days of actual usage before recommending a commitment mix, and continuously monitors utilization afterward so under-utilized commitments get flagged early.
