LOADING
PlatformHow it worksFinOps LeadershipPricingBlogEventsCalculatorFor FinanceFor EngineeringFor FinOpsAboutCareersContact Book a demo
19 November 2026

AWS Graviton Cost Savings: How to Migrate to Arm in 6 Steps

AWS Graviton can lower compute costs with better price performance. Follow a six-step migration plan, avoid common mistakes and prove the savings.

AWS Graviton Cost Optimization EC2

AWS Graviton Cost Savings: How to Migrate to Arm in 6 Steps

Quick answer: AWS Graviton processors are Arm-based chips designed by AWS. Graviton instances are typically priced below comparable Intel and AMD instances, and AWS claims up to 40% better price performance for suitable workloads. Most Linux workloads move with a rebuild and a benchmark, not a rewrite. The saving comes from the combination of a lower hourly price and more work per dollar.

Few AWS changes pay back as reliably as moving compute to Graviton. The architecture is mature, the managed services support it, and for a lot of teams the work amounts to rebuilding a container image and running a proper load test.

Yet plenty of fleets still sit on x86 out of habit. This guide explains where the money comes from, which services support Graviton, and a six-step migration that keeps risk low.

What is AWS Graviton?

Graviton is AWS's own family of server processors, built on the Arm architecture instead of x86. You recognise the instances by the “g” in the name: m7g, c7g, r7g, and so on. Graviton2, Graviton3 and Graviton4 are already in wide use, and AWS made its first Graviton5 instances, the M9g and M9gd, generally available in June 2026. AWS says they deliver up to 25% better compute performance than Graviton4-based instances.

Where do the savings come from?

Two separate numbers get mixed up in most Graviton discussions:

  1. Lower list price. A Graviton instance usually costs less per hour than the equivalent Intel or AMD instance with the same vCPU and memory.
  1. Better price performance. AWS claims up to 40% better price performance than comparable current-generation x86 instances. That's a vendor claim about suitable workloads, not a guarantee.

The first is a certainty you can read off the pricing page. The second depends on your code, which is why benchmarking matters. Confusing the two is the most common reason a migration disappoints: someone expects 40% and sees 15%, when 15% was a perfectly good result.

The real saving also depends on what you do with extra performance. If the same work runs on fewer or smaller instances, the bill falls faster than the price gap alone suggests.

Which AWS services support Graviton?

ServiceHow to switch
Amazon EC2Choose a Graviton instance type and an arm64 AMI
Amazon RDS and AuroraChange the DB instance class
Amazon ElastiCache and OpenSearch ServiceChange the node type
AWS LambdaSet the function architecture to arm64
Amazon EKS and ECSAdd Graviton nodes and multi-architecture images
AWS FargateChoose the ARM64 platform in the task definition

Managed services are the easiest starting point, because you don't touch application code. A database or cache node change is an instance class change, usually done in a maintenance window. Lambda is just as simple, and arm64 functions are billed at a lower duration price, as we cover in our Lambda cost optimization guide.

Will your workload run on Graviton?

Probably, with a few exceptions. The usual suspects are easy to check:

  • Interpreted and managed runtimes such as Java, Python, Node.js, Go, .NET and Ruby generally run unchanged.
  • Native dependencies like compiled libraries, drivers and agents need an arm64 build.
  • Container images must be built for arm64 or as multi-architecture images.
  • Third-party software needs a vendor-supported arm64 version.
  • Windows Server workloads generally aren't candidates, so plan those separately.

AWS provides an open-source Porting Advisor for Graviton that scans code and flags architecture-specific issues, which saves a lot of guesswork on larger codebases.

AWS Graviton migration overview: EC2, RDS and container workloads move to Arm for lower cost and better price performance

The 6-step Graviton migration plan

Step 1: Find the biggest x86 spend

Start from your bill, not from your architecture diagram. List compute spend by instance family and account, then rank. A handful of families usually carry most of the money. In AWS Compute Optimizer, EC2 recommendations can include Graviton options, which gives you a first shortlist. If you haven't right-sized yet, do that first: rightsizing shrinks the base you're migrating and keeps the savings honest.

Step 2: Check compatibility

Walk through the checklist above for each candidate workload. Run the Porting Advisor on code, check each dependency for an arm64 release, and confirm your monitoring and security agents support Arm. Anything blocked gets parked with a note, not argued about.

Step 3: Build for arm64

For containers, build multi-architecture images, for instance with Docker buildx, and push them under a single tag. The same deployment can then run on either architecture, which makes rollback trivial. For EC2 AMIs, bake an arm64 version alongside the x86 one. Don't forget your CI runners: if your pipeline only builds on x86, it will keep producing x86 artifacts.

Step 4: Benchmark with real traffic

Don't trust a generic benchmark. Take a representative service, replay production-like load, and compare cost per request (or per job) across both architectures at the same latency target. That one number tells you whether the migration is worth doing, and it's the number finance will understand. Our guide to AWS cost per transaction shows how to calculate it.

Step 5: Roll out gradually

Run a mixed fleet. Add a small share of Graviton capacity next to the x86 fleet, watch latency, error rates and CPU, then shift weight in stages. In Kubernetes, node selectors and node pools per architecture let you move services one at a time. Keep the x86 path alive until you've been stable for a full business cycle, including month-end or whatever your busiest period is.

Step 6: Re-size your commitments

This step is the one teams forget, and it can quietly eat the savings. Standard Reserved Instances and EC2 Instance Savings Plans are tied to an instance family, so moving off that family leaves them unused. Compute Savings Plans apply across instance families, so they follow the workload. Before you migrate a large fleet, check what commitments cover it, and re-plan afterwards. Our Savings Plans and Reserved Instances guide explains how, and our post on Savings Plan utilization shows how to spot waste early.

Graviton and Spot: a strong pairing

Graviton and Spot stack. Spot is a purchase option, Graviton is a processor choice, and the two don't conflict. Stateless services and CI runners on Graviton Spot nodes often get the deepest discount available for ordinary compute. Remember to diversify across Graviton generations and sizes so Spot has many pools to draw from.

Mistakes to avoid

  • Judging by hourly price alone. Compare cost per unit of work.
  • Migrating before right-sizing. You'll move waste to a new architecture.
  • Forgetting commitments. Stranded reservations can wipe out the gain.
  • Skipping the rollback path. Multi-architecture images make this free.
  • Testing on synthetic load. Production traffic shapes expose what benchmarks hide.

How to prove the savings

Tag migrated workloads, or group them by instance family in your cost data, and track three things before and after: total compute cost for the workload, cost per request or job, and p95 latency. If cost falls and latency holds, you have a clean story. A clear before-and-after on cost per transaction beats a percentage on a slide, and it's the simplest way to make FinOps value visible to the business.

Frequently asked questions

Is AWS Graviton cheaper than Intel or AMD?

Typically, yes. Graviton instances usually list below comparable x86 instances, and AWS claims up to 40% better price performance for suitable workloads. Your actual result depends on your application, so benchmark it.

Do I need to rewrite my application for Graviton?

Usually not. Applications in Java, Python, Node.js, Go and .NET generally run unchanged. The work is in rebuilding for arm64 and replacing any x86-only native dependencies.

Is Graviton only for new workloads?

No. Many teams migrate existing production services in stages, starting with stateless services and managed databases and caches.

What's the difference between Graviton4 and Graviton5?

Graviton5, available in M9g and M9gd instances, is the newer generation. AWS says it delivers up to 25% better compute performance than Graviton4-based instances. Regional availability varies, so check your region before planning.

Final thoughts

Graviton rewards the unglamorous work: an inventory, a compatibility check, a multi-architecture build and an honest benchmark. Do those four things and the migration becomes routine. Keep an eye on commitments, measure cost per unit of work, and the savings will stay on the bill instead of leaking out through unused reservations.

Want to see which of your workloads offer the biggest return? Book a demo with TruCost.Cloud and we'll look at your own AWS data together.

Related reading

See your own bill allocated to 100%

Thirty minutes with a FinOps engineer, on your AWS data if you like. No slideware.

Book a demo