Usually 20% cheaper, sometimes not worth it
Is AWS Graviton actually cheaper?
Graviton is AWS’s own Arm processor family. The pricing claim is real, the performance claim is usually real, and the migration effort is the part that decides whether it is worth your time.
Yes. Graviton instances list roughly 20% below the equivalent x86 instance, and AWS claims up to 40% better price-performance for suitable workloads. A t4g.medium is about $0.0336 an hour against $0.0416 for a t3.medium, so about 19% less for the same vCPU and memory. The saving is real and immediate for anything running on an interpreted or JIT language; it is a project for anything with compiled native dependencies.
Like-for-like hourly prices
| x86 | $/hour | Graviton | $/hour | Difference |
|---|---|---|---|---|
| t3.medium | $0.0416 | t4g.medium | $0.0336 | −19% |
| t3.large | $0.0832 | t4g.large | $0.0672 | −19% |
| m5.large | $0.096 | m6g.large | $0.077 | −20% |
| m5.xlarge | $0.192 | m6g.xlarge | $0.154 | −20% |
| c5.large | $0.085 | c6g.large | $0.068 | −20% |
| r5.large | $0.126 | r6g.large | $0.1008 | −20% |
| db.m5.large (RDS) | varies | db.m6g.large | varies | ~−10% |
Representative on-demand Linux prices. The gap is consistent at roughly 20% for EC2 and rather smaller for RDS.
When it is nearly free to switch
- →Managed services. RDS, ElastiCache, OpenSearch and MemoryDB all offer Graviton instance classes and the switch is a modify operation. Nothing in your application changes because nothing in your application knows.
- →Lambda. A single arm64 setting in the function configuration. For interpreted runtimes with no native dependencies it is a one-line change for around 20% off, and Lambda also runs slightly faster on Graviton, which reduces the duration charge as well.
- →Containers on interpreted runtimes. Node, Python, Ruby and JVM images generally build for arm64 without incident. If your base images publish multi-arch manifests, this is a build-and-deploy rather than a port.
When it is a project
- →Compiled native dependencies. Anything with C extensions, custom binaries or vendored native libraries needs rebuilding and retesting on arm64. Usually possible, occasionally blocked by a dependency that never shipped an Arm build.
- →Commercial software with x86-only licensing. Some vendors simply do not support Arm, and some licence per architecture. Check before planning the migration rather than after.
- →CI that only builds x86. You need multi-arch builds in the pipeline before you can deploy multi-arch, and that work lands on whoever owns CI. It is worth doing anyway; it is not free.
- →Existing x86 Savings Plans and RIs. A Compute Savings Plan covers both architectures, so migrating is fine. EC2 Instance Savings Plans and standard RIs are tied to an instance family, so migrating away from x86 while committed means paying for a commitment you no longer use.
Do this before migrating anything
Rightsize first. A 20% discount on an instance twice the size it needs to be is a worse outcome than the correct size on x86. The right order is: remove what is idle, size what is left correctly, migrate the survivors to Graviton where it is cheap to do so, then commit to the baseline that remains.
Put numbers on it
EC2 idle cost estimator
What your under-5%-CPU instances cost, and what rightsizing the quiet ones saves.
Open calculator →
Savings Plans calculator
On demand against one and three year commitments, and why rightsizing has to come first.
Open calculator →
Graviton savings calculator
About 20% off the equivalent x86 instance, applied to the share of your fleet you can realistically port.
Open calculator →
Frequently asked questions
Is Graviton actually faster, or just cheaper?
For most general-purpose workloads Graviton3 and Graviton4 are competitive on raw throughput and better on performance per pound, which is the number that matters. Single-thread-heavy workloads and anything tuned to specific x86 vector instructions can be slower. Benchmark your own workload rather than trusting either the marketing or the scepticism.
Do Savings Plans cover Graviton?
Compute Savings Plans cover any instance family, size, region and architecture, so they cover Graviton automatically. EC2 Instance Savings Plans and standard Reserved Instances are locked to a family, so they do not follow you from m5 to m6g. This is a good reason for smaller accounts to prefer Compute Savings Plans, which trade a slightly smaller discount for the freedom to change your mind.
Can I run mixed x86 and Graviton in one cluster?
Yes, both ECS and EKS support mixed-architecture node groups, and it is the sensible way to migrate: add Graviton capacity, shift a portion of traffic, watch, then continue. It does require multi-arch container images.
How much would this save on a typical bill?
If half your compute is easily portable, a 20% saving on half of your compute spend is 10% of that spend. On a $10,000 a month bill where compute is 60%, that is around $600 a month. Worth planning, rarely worth doing before the idle instances are gone.
Keep reading
Prices checked against AWS list rates on 2 August 2026. AWS changes prices; treat every figure here as a close approximation rather than a quote.
£499 fixed. Free scan first. 20%+ found or it’s free.
The scan is read-only — a role you create and delete, no keys shared — and shows your estimated monthly saving before anyone talks about money.