Azure Reservations vs AWS Savings Plans vs Google CUDs

Par Rohit Athaide · Publié le 4 nov. 2026

Short answer: each cloud gives a lower price in return for a commitment. The programs come in two shapes, a commitment to a specific resource and a commitment to hourly spend. The first gives a bigger discount and less freedom. The second gives a smaller discount and keeps you flexible. Which fits depends on how stable your workload is.

Key points

• Resource-based commitments give the larger discount and the least flexibility.
• Spend-based commitments keep you free to change machine type and, on some clouds, region.
• Coverage is mostly compute. Storage and bandwidth are usually outside.
• You pay the commitment whether or not you use it.
• Do not commit for workloads that may move, shrink or stop.

What are the two shapes of commitment?

1. Commit to a resource. You agree to a specific machine type in a region for one or three years. The discount is larger, and changes are limited.
2. Commit to spend. You agree to spend a set amount per hour on a family of services. The discount is a little smaller, and you can change what you run.

Names and rules change, so read the vendor’s page before you buy. Reserved instances and savings plans explains the mechanics.

What does a worked example look like?

Made-up round numbers to show the method, not real prices or real discount levels. A steady server costs 100 per month on demand, and you run 10 of them.

Now the workload shrinks to 6 servers after four months. Under the 3-year resource commitment you still pay for 10: 500 per month, while the on-demand equivalent would cost 600 for six, so the commitment still wins by 100. If it shrinks to 3 servers, on demand costs 300 and the commitment is 200 worse. The rule: commit only to the base load that you are confident will run for the whole term, and leave the peaks on demand.

What should I ask about each program?

• What exactly is covered? Some cover only virtual machines. Others include containers and functions.
• Can I change it? Can you switch machine families, regions or operating systems?
• Can I cancel or sell it? Some allow exchange or resale. Others do not.
• How do I pay? All upfront, partly upfront or monthly, which affects the rate.
• What happens when the term ends? Usually back to the on-demand rate.
• How do discounts stack? On Google Cloud, resource-based commitments apply first, then flexible ones, then automatic discounts. Check the current rules.

When should I not commit?

Do not commit for workloads you may shut down, move or resize within the term. Use spot instances for flexible batch work. A commitment on a machine you retire is money spent on nothing. Commitments also do not cover data transfer, so they do not fix an egress-heavy bill.

Which mistakes are common?

• Committing to peak load instead of base load.
• Buying a resource commitment for a machine family you plan to leave.
• Forgetting the term end date, then paying on-demand rates.
• Comparing a committed price on one cloud with a list price on another.

Decision checklist

• What is the lowest steady usage over the last three months?
• Will the workload run for the whole term?
• Is flexibility worth a smaller discount?
• Is there a calendar reminder for the term end?

Next steps

Our estimates use list prices, so you can compare services before discounts. Discounts depend on your commitments and are applied as a separate step. See list prices vs what you actually pay and cloud pricing models explained.