AWS has invested billions in its custom silicon programme — Graviton CPUs, Inferentia and Trainium AI chips, and the Nitro hypervisor architecture — and the resulting price-performance advantages are substantial. For enterprise buyers, Graviton is not just a technical choice: it is one of the most accessible levers for reducing AWS compute costs without renegotiating a single contract clause.
This article is part of our AWS Enterprise Negotiation Guide, which covers the full range of commercial strategies for organisations spending $1M+ annually on AWS. For context on how Graviton fits into your broader AWS cost strategy, see our AWS cost optimisation strategies overview.
Before exploring migration tactics, it is also worth understanding how Graviton savings interact with your EDP discount. Our AWS EDP negotiation playbook explains how AWS account teams respond to buyers who demonstrate strong Graviton adoption rates — typically with greater commercial flexibility. See also our AWS Savings Plans vs Reserved Instances guide for how to stack commitment-based discounts on top of Graviton's inherent price-performance advantage.
AWS Graviton3 instances are priced approximately 20% lower than equivalent x86 instances for the same compute capacity — before Reserved Instance or Savings Plan discounts. Combined with RI or Savings Plan commitments, total savings versus on-demand x86 can reach 65–70%. Organisations migrating 50%+ of eligible compute to Graviton frequently report seven-figure annual savings.
What Is AWS Graviton?
AWS Graviton is Amazon's family of custom ARM-based processors, designed in-house and optimised for cloud-native workloads running on EC2. The programme has evolved through three generations: Graviton (2018), Graviton2 (2020), and Graviton3 (2022), with Graviton4 introduced in 2024 for select instance families.
AWS's custom silicon strategy goes beyond Graviton. The broader portfolio includes Inferentia (ML inference), Trainium (ML training), and the Nitro System (virtualisation and security offload). Each chip family is purpose-built for a specific workload class, delivering superior efficiency compared to equivalent third-party silicon at lower per-unit cost to AWS — savings that are partially passed through to customers in the form of lower instance prices.
Graviton Generation Comparison
| Generation | Launch Year | Architecture | Key Improvement | Representative Instance |
|---|---|---|---|---|
| Graviton | 2018 | ARM Cortex-A72 | First-gen custom, limited workload fit | A1 |
| Graviton2 | 2020 | Custom Neoverse N1 | 7× performance vs Graviton1, 40% better price-perf vs x86 | M6g, C6g, R6g, T4g |
| Graviton3 | 2022 | Custom Neoverse V1 | 25% better performance vs Graviton2, 60% better ML | M7g, C7g, R7g, Hpc7g |
| Graviton4 | 2024 | Custom Neoverse V2 | 30% better performance vs Graviton3, higher core density | M8g, C8g, R8g |
Graviton Pricing vs x86: The Numbers
The price differential between Graviton and equivalent x86 instances is the most direct savings lever available without requiring any negotiation or commitment change. AWS publishes on-demand pricing for both families, and the Graviton advantage is consistent across regions and instance sizes.
On-Demand Price Comparison (us-east-1, March 2026)
| Workload Type | x86 Instance | On-Demand (x86) | Graviton Instance | On-Demand (Graviton) | Savings |
|---|---|---|---|---|---|
| General purpose (8 vCPU) | m7i.2xlarge | $0.4032/hr | m7g.2xlarge | $0.3264/hr | 19% |
| Compute optimised (8 vCPU) | c7i.2xlarge | $0.3570/hr | c7g.2xlarge | $0.2890/hr | 19% |
| Memory optimised (8 vCPU) | r7i.2xlarge | $0.6720/hr | r7g.2xlarge | $0.5376/hr | 20% |
| Burstable (2 vCPU) | t3.large | $0.0832/hr | t4g.large | $0.0672/hr | 19% |
These on-demand differentials are substantial on their own. But the real compounding effect comes when you layer Graviton's inherent price advantage with Reserved Instance or Savings Plan commitments — which apply the same percentage discount to a lower base price.
Savings Stacking: Graviton + Commitments
| Scenario | Hourly Rate (m7i.2xlarge equivalent) | Monthly Cost (720 hrs) | vs On-Demand x86 |
|---|---|---|---|
| On-Demand x86 (baseline) | $0.4032 | $290 | — |
| On-Demand Graviton | $0.3264 | $235 | -19% |
| 1-yr RI x86 (no upfront) | $0.2600 | $187 | -35% |
| 1-yr RI Graviton (no upfront) | $0.2100 | $151 | -48% |
| 3-yr RI Graviton (all upfront) | $0.1380 | $99 | -66% |
For most Java, Python, Node.js, and Go workloads, Graviton migration is largely a recompile-and-test exercise. For containerised environments, migration is even simpler — AWS provides multi-architecture container image support natively in ECR. Organisations frequently underestimate how straightforward the transition is for modern application stacks.
Which Workloads Migrate to Graviton?
Graviton is not universally suitable for every workload — but the range of compatible workloads is far broader than most organisations initially assume. The key constraint is the ARM instruction set architecture: workloads that depend on x86-specific binaries, Windows Server (Graviton is Linux-only), or certain legacy compiled libraries require additional assessment.
Graviton Workload Compatibility Matrix
| Workload Type | Graviton Compatibility | Migration Effort | Notes |
|---|---|---|---|
| Java 11+ applications | Excellent | Low — JVM abstracts architecture | AWS Corretto optimised for Graviton |
| Python 3.x workloads | Excellent | Low — interpreted language | NumPy/SciPy have ARM builds |
| Node.js / Go / Rust | Excellent | Low — recompile required | Native ARM support in all modern versions |
| Containerised microservices | Excellent | Low — multi-arch images | ECS, EKS support ARM natively |
| Linux-based databases (MySQL, PostgreSQL) | Good | Low — RDS/Aurora Graviton supported | Aurora Graviton delivers ~35% better price-perf |
| .NET Core / .NET 6+ | Good | Medium — framework version dependent | .NET 6+ has ARM64 support |
| Windows Server workloads | Not supported | N/A | Graviton is Linux-only |
| x86-compiled legacy binaries | Poor | High — rewrite or emulation | Emulation (Rosetta-style) not available on EC2 |
| ML inference (non-GPU) | Excellent | Low — use Inferentia instead | Consider Inferentia for cost-optimised inference |
AWS Inferentia and Trainium: AI/ML Custom Silicon
Beyond Graviton, AWS has developed two purpose-built AI chips that deliver significant cost advantages for machine learning workloads compared to GPU-based alternatives. Understanding these options is increasingly important as AI/ML spend grows within enterprise AWS estates.
Inferentia (Inf2)
AWS Inferentia is designed for high-throughput, low-latency ML inference. The Inf2 instance family uses Inferentia2 chips, which deliver up to 4× higher throughput and 10× lower latency than GPU-based inference at a fraction of the cost. For organisations running high-volume inference workloads — recommendation engines, fraud detection, NLP APIs — migrating to Inferentia can reduce ML inference costs by 30–70% versus equivalent GPU instances.
Trainium (Trn1)
AWS Trainium is optimised for ML training workloads. Trn1 instances deliver up to 50% cost savings compared to GPU-based training for supported frameworks (PyTorch, TensorFlow, JAX). AWS offers Trainium under Savings Plans, with 1-year and 3-year discount options that compound the base price advantage further.
When negotiating your EDP or private pricing arrangements, committing to Graviton and custom silicon adoption targets gives AWS account teams a business case to take to their internal approval chain for enhanced discounts. Enterprises that have agreed measurable Graviton adoption milestones as part of EDP renegotiations have secured 2–5 percentage points of additional top-line discount — worth $100K–$500K annually at mid-market spend levels.
Migration Strategy: A Phased Approach
A systematic migration to Graviton does not require a big-bang transformation. The most successful enterprise Graviton programmes use a phased approach that builds confidence, captures quick wins, and progressively expands the Graviton footprint over 12–18 months.
Phase 1: Discovery and Assessment (Weeks 1–4)
Begin by inventorying your existing EC2 fleet with a focus on identifying Graviton-compatible candidates. AWS provides tools including Compute Optimizer and the Migration Evaluator that automatically flag instances eligible for Graviton migration and estimate potential savings. Focus on instances running Linux-based workloads with modern language runtimes first — these are your easiest wins.
Phase 2: Pilot Migration (Weeks 5–12)
Migrate 10–20% of eligible workloads to Graviton instances in a non-production environment. Focus on stateless services with automated rollback capability. Measure performance against your x86 baseline using your existing monitoring stack. For Java workloads, switch to AWS Corretto 11+ for optimal Graviton performance.
Phase 3: Production Rollout (Months 4–9)
With pilot performance data validated, begin systematic production migration. Update your Auto Scaling launch templates and ECS/EKS task definitions to specify Graviton instances. For containerised environments, rebuild container images with multi-architecture support using Docker BuildKit or AWS CodeBuild's ARM build support.
Phase 4: Reserved Capacity Commitment (Months 6–12)
Once Graviton accounts for a meaningful share of your compute estate, convert on-demand Graviton usage to Reserved Instances or Savings Plans. This is the point at which the compound savings are maximised. Coordinate this with your Savings Plans vs RI strategy — and time it ahead of your EDP renewal to demonstrate to AWS that your architectural efficiency is improving, which strengthens your negotiating position.
Graviton and Your EDP: The Commercial Angle
AWS account teams receive internal credit for helping customers adopt custom silicon — it demonstrates AWS's differentiated technology position and builds customer stickiness. This internal incentive structure means that a credible Graviton migration roadmap has direct commercial value in EDP negotiations.
When entering an EDP renegotiation or new EDP discussion, position your Graviton adoption programme as a bilateral commitment: you will move workloads to Graviton in exchange for enhanced EDP discounts or additional service credits. AWS is typically willing to provide introductory credits for Graviton migration work, architectural support from Solutions Architects, and marginal improvements to top-line EDP discount in exchange for concrete adoption milestones.
For a complete negotiation framework, see our AWS EDP negotiation playbook. For broader cloud commitment strategy, our cloud commitment strategy guide explains how to coordinate Graviton RI commitments with your overall portfolio approach.
10 Tactics for Maximising Graviton Savings
Common Mistakes and How to Avoid Them
The most frequent mistake organisations make with Graviton is over-engineering the assessment phase. Teams spend months building detailed compatibility matrices when a simple pilot migration of 5–10 non-critical services would answer the same questions in two weeks at near-zero risk.
A second common error is failing to update Reserved Instance purchasing after migrating to Graviton. Organisations that migrate EC2 instances from x86 to Graviton but still hold x86 RIs face wasted commitment spend unless those RIs can be converted or exchanged. When planning Graviton migrations, always audit your existing RI portfolio and plan conversions accordingly. See our RI vs Savings Plans guide for how to structure flexible commitments that accommodate architectural change.
Third, many teams focus exclusively on EC2 and miss the managed service opportunity. Graviton-based RDS, Aurora, ElastiCache, and OpenSearch instances deliver the same 20% cost advantage without any application-level changes. For database-heavy environments, this is often the highest-value Graviton opportunity available.
ROI Model: 500-Instance Enterprise Example
Consider an organisation running 500 Linux EC2 instances, 70% of which are Graviton-compatible general-purpose instances (m-family), currently on x86. Monthly on-demand cost at $0.40/hr per instance: approximately $144,000/month ($1.73M/year).
After migrating 350 instances to Graviton equivalents (m7g) at $0.326/hr and purchasing 1-year no-upfront RIs at $0.21/hr: monthly cost falls to approximately $77,000/month for those instances ($924K/year). The remaining 150 x86-only instances maintain their existing commitment structure. Total annual saving on compute alone: approximately $800K — before EDP discount enhancements that Graviton adoption may unlock.
Need Help Modelling Your Graviton Savings Opportunity?
Our advisory partners have helped enterprise AWS customers capture £500K–£5M+ in annual savings through systematic Graviton migration programmes combined with EDP optimisation.
Get a Free Assessment →Graviton in the Context of Your AWS Commercial Strategy
Graviton is most valuable when embedded in a broader AWS commercial strategy rather than treated as a standalone cost-cutting exercise. The organisations that extract the most value from Graviton are those that treat it as an architectural programme with commercial implications — not just a technical migration project.
In practice, this means coordinating Graviton adoption milestones with your EDP renewal calendar (see our software renewal timing strategy), using adoption evidence to unlock enhanced Marketplace private offers (see our AWS Marketplace private offers guide), and incorporating Graviton-based compute projections into your multi-year AWS spend forecasting.
For the full AWS enterprise negotiation picture, return to our AWS Enterprise Negotiation Guide or explore our rankings of the best AWS negotiation consulting firms if you need specialist advisory support for your next AWS commercial review.
Enterprise buyers looking for a broader market perspective on what is achievable in cloud contract negotiations should also review our Cloud Negotiation Playbook white paper, which covers AWS, Azure, and GCP in depth.
Frequently Asked Questions
How much cheaper are Graviton instances than x86 EC2 instances?
Graviton instances are typically 19–20% cheaper than equivalent x86 instances on an on-demand basis. Combined with Reserved Instance or Savings Plan discounts, total savings versus on-demand x86 can reach 60–70%.
Does Graviton support Windows Server?
No. AWS Graviton is a Linux-only platform. Windows Server workloads must remain on x86-based instance families. If a significant proportion of your estate is Windows, your Graviton opportunity is proportionally smaller.
Can I use Graviton with my existing Reserved Instances?
Existing x86 RIs cannot be directly applied to Graviton instances. However, Convertible RIs can be exchanged for Graviton equivalents. Savings Plans (Compute Savings Plans) apply to any EC2 instance regardless of architecture, making them more flexible for organisations planning Graviton migrations.
Will Graviton adoption help in my EDP negotiation?
Yes. AWS account teams are incentivised to drive Graviton adoption as part of demonstrating AWS's differentiated value. A credible Graviton adoption roadmap — ideally with measurable milestones — provides a commercial basis for requesting enhanced EDP discounts, migration credits, or additional Solutions Architect support.