An AWS Spot Instance is spare EC2 capacity sold at 60 to 90 percent off the on demand rate, reclaimable by AWS on 2 minutes notice. The vehicle is built for fault tolerant, stateless, or checkpoint friendly workloads such as batch processing, big data analytics, CI runners, and ephemeral container nodes. Across 240 enterprise AWS accounts benchmarked in 2024 and 2025, the median spot discount was 71 percent off on demand, with GPU instance families in low demand Availability Zones often reaching 88 percent.
Definition
AWS Spot Instance: Unused EC2 capacity that AWS sells at a steep discount to the on demand rate, with the right to reclaim the instance on 2 minutes notice. Pricing is set by AWS using internal capacity signals and is no longer the auction model of the early years. Spot complements Reserved Instances and Savings Plans in a mature commitment portfolio. Sits inside the broader AWS EDP framework.
Spot is not a commitment vehicle. There is no minimum spend, no term, and no upfront. The trade is that AWS reclaims the instance when capacity tightens, signalling the reclaim through a 2 minute interruption notice on instance metadata. Workloads must catch the signal, checkpoint state, and drain gracefully. Modern services that handle this well include EMR, EKS with Karpenter, ECS with capacity providers, and Batch.
The savings stack with other commitments. A workload running on a 3 year All Upfront Compute Savings Plan reservation does not get the spot discount on top. But a workload running on bare spot capacity above the commitment floor gets the full spot discount. Mature FinOps programs layer spot fleets on top of Savings Plan coverage to capture both. See the cloud cost management definition for the savings ladder context.
Benchmark your spot usage
We benchmark spot coverage, interruption rates, and effective discount realization across 240 enterprise AWS accounts. Send the spot fleet profile and we return optimization intelligence in 48 hours.
Discount benchmark by family
From our 2024 and 2025 sample of enterprise AWS accounts: general purpose families (m5, m6i, m7i) ran 60 to 78 percent discount, compute optimized (c5, c6i) ran 64 to 82 percent, memory optimized (r5, r6i) ran 58 to 76 percent, and GPU families (g4dn, g5, p4d) ran 70 to 90 percent. Interruption rate averaged 4 to 12 percent monthly across the sample, with diversified instance fleets running below 6 percent. Methodology: NDA invoice data, deal size brackets 500K to 80M annually.
For the surrounding vocabulary, see the Reserved Instance definition, the Savings Plan definition, the consumption based pricing definition, and the glossary hub. For AWS benchmarks, see the cloud infrastructure benchmarks and the AWS pricing profile.
Frequently asked questions
What is an AWS Spot Instance?
An AWS Spot Instance is spare EC2 compute capacity that AWS sells at 60 to 90 percent off the on demand rate. AWS can reclaim the instance on 2 minutes notice when capacity is needed elsewhere. Spot is built for fault tolerant, stateless, or checkpoint friendly workloads.
How big is the spot discount?
Across 240 enterprise AWS accounts benchmarked in 2024 and 2025, the median spot discount was 71 percent off on demand, with the inter quartile range running 63 to 84 percent. GPU spot instances often run at the highest discount in low demand AZs.
What workloads should run on spot?
Stateless web tiers behind autoscaling, batch ETL, big data clusters with checkpoint and recovery, CI runners, container nodes, machine learning training with checkpoint resumption, video transcoding, and Monte Carlo simulations. Avoid stateful databases or workloads with strict deadlines under 2 minutes.
Benchmark your AWS contract
Send us the AWS EDP proposal, RI portfolio, or spot fleet profile. We return discount and unit price intelligence in 48 hours.
Benchmark your contract against modelled deal cohorts, or decode an agreement free in about a minute.
Decode a contract free →