Spot & Preemptible VM Benchmarks Across AWS, Azure, GCP, and OCI

1. The Hidden Volatility of Hyperscaler Spot Pricing

Engineering teams adopt Spot VMs (AWS EC2 Spot, Azure Spot VMs, Google Cloud Spot VMs) to reduce compute costs for stateless microservices, CI/CD runners, and data processing pipelines. However, on AWS and Azure, Spot pricing fluctuates dynamically based on regional supply and demand, and base On-Demand list prices start 2x higher than Oracle Cloud Infrastructure (OCI).

Even more problematic is rigid instance sizing: if your batch container needs 6 vCPUs and 16 GB of RAM, AWS and Azure force you into an 8-vCPU / 32-GB instance family (`m7i.2xlarge` or `D8s_v5`), wasting 25% to 50% of the memory you pay for.

2. OCI Preemptible Instances: Flat 50% Discount + Custom Core/RAM

On Oracle Cloud Infrastructure, Preemptible Instances apply a fixed 50% discount directly to `VM.Standard.E5.Flex` (AMD EPYC Genoa), `VM.Standard3.Flex` (Intel Xeon), and `VM.Standard.A1.Flex` (Ampere Altra ARM) shapes.

Because 1 OCI OCPU equals 2 hardware execution threads (2 vCPUs on x86) and memory is billed independently at $0.0015/GB-hour ($0.00075/GB-hour preemptible), an 8-vCPU (4 OCPU) / 32-GB RAM Linux Preemptible VM on OCI costs roughly $64/month, compared to $85-$140/month on AWS/Azure Spot and over $280-$338/month On-Demand.

3. Combining OKE Node Pools with Mixed On-Demand & Preemptible Shapes

In OCI Container Engine for Kubernetes (OKE), we architect dual node pools: a baseline On-Demand `E5.Flex` pool for stateful ingress and core APIs, paired with an auto-scaling Preemptible `E5.Flex` or `A1.Flex` pool that handles burst traffic, document OCR, and ETL workers with graceful 2-minute `SIGTERM` pod draining.