EC2 vs Azure VM vs Compute Engine: instance equivalents

Von Rohit Athaide · Veröffentlicht am 4. Nov. 2026

Short answer: every cloud names its virtual machines differently, so match by what the machine does, not by what it is called. Four facts decide equivalence: vCPU count and what a vCPU is, memory per vCPU, CPU type, and whether performance is fixed or burstable.

Key points

• Names pack a family, a generation and a size in different orders on each cloud.
• General purpose is about 4 GB per vCPU, compute-heavy about 2 GB, memory-heavy about 8 GB or more.
• Never match a burstable VM against a fixed-performance one.
• Arm and x86 are not interchangeable without testing.
• Match the generation, disk and network too.

How do I read a size name?

Most names pack the same facts in different orders:

• A family letter for the purpose: general purpose, compute-heavy, memory-heavy, burstable.
• A generation number.
• A size that sets vCPU count and memory.
• Optional suffixes for Arm processors, local disks or faster networking.

AWS m or c families, Azure’s D, F and E series, Google’s n2 or e2 families and Oracle’s flexible shapes all follow this pattern in their own way. Vendor lists: EC2 instance types, Azure VM sizes, Compute Engine machine families and OCI compute shapes.

What four facts decide equivalence?

1. vCPU count and what a vCPU is. See what is a vCPU.
2. Memory per vCPU. General purpose is usually about 4 GB per vCPU. Look up each family to be sure.
3. CPU type. Arm and x86 are not interchangeable without testing. See Arm or x86.
4. Performance model. Burstable machines earn credits for short peaks. Fixed-performance ones give steady speed. See burstable vs fixed-performance.

What does a worked example look like?

Made-up names and numbers to show the method. You need a steady web server with 4 vCPUs and 16 GB, on x86.

Cloud C looks half the price and is not an equivalent, because it is burstable. Putting it in the table would make the other two look expensive. This is the most common error in size comparisons.

Which traps catch people?

• Matching by letter. A letter means different things on different clouds.
• Ignoring generation. A newer generation of the same size often costs the same or less and runs faster.
• Ignoring disk and network. A local disk or a faster network card is part of the size.
• Comparing across regions. Prices move with location.
• Comparing a shape you can resize in small steps with a fixed size.

How do our estimates choose?

Our estimates pick a fixed-performance, general-purpose size on each cloud with the same vCPU and memory, and record the reason in the service map. The virtual machine compare page shows the sizes and price lines used.

Decision checklist

• Same vCPU count and same definition of vCPU?
• Same memory per vCPU?
• Same CPU architecture?
• Both fixed-performance, or both burstable?
• Same region group and same disk type?

Next steps

Choosing a VM size covers sizing from your own load. The web app estimator prices the servers on all four clouds. Windows vs Linux VM cost covers the licence difference you also need to match.