Reserved Instances & Committed Use Discounts Explained (2026)

AWS offers two commitment models. Reserved Instances (RIs) lock you into a specific instance type, OS, and region for 1 or 3 years. Standard RIs offer higher discounts but less flexibility; Convertible RIs let you change instance families but at lower discounts. Savings Plans are the newer, more flexible option — you commit to a dollar amount per hour of usage rather than specific instances. Compute Savings Plans apply across any instance family, region, or OS, while EC2 Instance Savings Plans are locked to a family and region but offer deeper discounts.

Example: An m5.xlarge costs $0.192/hr on-demand in us-east-1. With a 1-year Compute Savings Plan (no upfront), you pay $0.121/hr — a 37% saving. With a 3-year all-upfront RI, the effective rate drops to $0.079/hr — 59% off. For most workloads, Savings Plans are the recommended starting point because they provide flexibility if you change instance families mid-term.

Azure Reserved VM Instances provide up to 72% savings over pay-as-you-go pricing. You commit to a specific VM series in a specific region for 1 or 3 years. Azure provides instance size flexibility within the same series — for example, a D2s_v3 reservation can apply to D4s_v3 at half utilization. Azure also offers Azure Savings Plan for Compute, which works similarly to AWS Savings Plans: you commit to a per-hour spend amount that applies flexibly across VM series and regions.

Example: A D4s_v5 (4 vCPU, 16 GB) costs $0.192/hr on-demand in East US. A 1-year reservation brings it to $0.121/hr (37% off). A 3-year all-upfront reservation drops to $0.077/hr (60% off). Azure also allows cancellation with an early termination fee (12% of remaining balance), giving some flexibility if your needs change.

GCP Committed Use Discounts (CUDs) offer up to 57% savings for 1-year commitments and up to 70% for 3-year commitments on specific machine families. Unlike AWS and Azure, GCP CUDs are always paid monthly — there is no upfront payment option. CUDs are tied to a specific machine family and region. GCP also offers Sustained Use Discounts (SUDs), which automatically apply up to 30% savings when instances run for more than 25% of the month — no commitment required. This is unique to GCP.

Example: An n2-standard-4 (4 vCPU, 16 GB) costs $0.194/hr on-demand in us-central1. Sustained Use Discounts bring it to $0.136/hr after a full month (30% off, automatic). A 1-year CUD brings it to $0.122/hr (37% off). A 3-year CUD reduces it to $0.087/hr (55% off). The fact that SUDs apply automatically makes GCP competitive even without explicit commitments.

Oracle Cloud Infrastructure (OCI) uses Universal Credits for commitment-based pricing. You commit to a minimum annual spend and receive a discount on the per-unit rate, typically 20-50% depending on the commitment size and negotiated terms. Oracle also offers capacity reservations that guarantee instance availability in a specific availability domain without any usage commitment.

Example: An Oracle E4 Flex shape at 4 OCPU / 16 GB costs about $0.10/hr on-demand. With a Universal Credits agreement, you can negotiate this down by 30-50%. Oracle's Always Free tier also includes 2 AMD VMs (1 OCPU, 1 GB each) and 4 Arm VMs (24 GB total) at no cost, permanently — making it the most generous free compute offering among all 8 providers.

DigitalOcean, Hetzner, Linode, and Vultr do not offer traditional reserved instance pricing. Their strategy is different: baseline prices that are already dramatically lower than the Big 3's on-demand rates. A comparable 4 vCPU / 16 GB droplet on DigitalOcean costs $48/mo ($0.071/hr) — about 57% less than an on-demand m5.xlarge on AWS. Hetzner's CX41 (4 vCPU, 16 GB) costs roughly €14/mo (~$15/mo) — which is 90% below AWS on-demand.

When to choose a smaller provider over a Big 3 reservation: If your workload doesn't require provider-specific services (like DynamoDB or BigQuery), using a smaller provider at on-demand pricing can be cheaper than the deepest Big 3 reservation discounts. The tradeoff is fewer regions, smaller service ecosystems, and less enterprise support.

Use this decision framework:

• Locked into AWS/Azure/GCP ecosystems? Use Savings Plans (AWS/Azure) for flexibility or RIs/CUDs for maximum discount on stable workloads.

• Variable workloads on GCP? Let Sustained Use Discounts work automatically and layer CUDs only on your baseline.

• Budget-sensitive startup or side project? Hetzner or DigitalOcean on-demand is likely cheaper than any Big 3 reserved pricing.

• Enterprise with negotiating power? Oracle Universal Credits can be very competitive if you commit enough annual spend.

• Mixed strategy? Run your core services on a Big 3 provider with commitments, and offload simple compute (CI/CD, staging, static hosting) to Hetzner or Vultr.

Start with on-demand to understand your usage patterns before committing. Use cloud provider tools (AWS Cost Explorer, Azure Advisor, GCP Recommender) to identify reservation opportunities. For variable workloads, prefer Savings Plans over RIs for flexibility. Mix commitment levels: commit 60-70% of your baseline usage and use on-demand or spot for peaks. Review commitments annually and adjust as your needs change. Consider convertible reservations if your architecture may change during the commitment period.

For smaller providers, the optimization is different: rightsize your droplet/server and consider annual billing if available (some providers offer 10-20% off for yearly payment).