Amazon EC2 M9g Instances

Airbnb is a leading travel company offering unique stays, experiences, and services that make it possible for guests to connect with communities in a more authentic way.

"AWS Graviton5-based Amazon EC2 instances are some of the fastest EC2 instances we have ever tested. In our performance tests, conducted using Airbnb's production search workloads, we are seeing improvements of up to 25% over other system architectures of the same generation, and up to 20% compared to prior generation Graviton4 instances. We are especially impressed with P95 latency for our critical workloads, helping to provide a consistent experience for Airbnb guests and hosts."

Denis Sheahan, Principal Performance Engineer, Airbnb

"Running EDA in the cloud allows us to scale compute to match the complexity of modern silicon design while improving engineering efficiency. AWS Graviton5, powered by Arm Neoverse V3, delivers up to 40% speedup for Arm’s own design and verification flows. That performance helps us bring the next generation of Arm processors to market faster and showcases the virtuous cycle of innovation between Arm and AWS.”

Karima Dridi, VP Productivity Engineering, Arm

“Atlassian has migrated more than 3000 EC2 instances for Jira and Confluence to AWS Graviton4-based instances. In our testing of Jira on AWS Graviton5-based M9g instances, we observed 30% higher performance and 20% lower latency compared to the prior generation and we look forward to AWS Graviton5 general availability.”

Paulo Almeida, Principal Site Reliability Engineer, Atlassian

"As AI-driven query volumes grow, performance and cost efficiency are inseparable. Graviton3 delivered a 25% query performance improvement over equivalent x86-based instances, and Graviton4 added a 64% further increase in queries per second. Once established on Graviton3, migrating to Graviton4 - and testing on Graviton5 - required no code changes. Early benchmarking of Graviton5-based instances shows a 36% performance boost and 16% higher concurrency compared to Graviton4. The migration to Graviton-based instances has delivered meaningful cost savings, which we reinvest in providing better service for our customers."

Alexey Milovidov, Founder and Chief Technology Officer, ClickHouse

"CockroachDB is a distributed SQL database built for enterprises that demand consistency, availability, and massive horizontal scalability. As a long-standing Graviton customer, we've live migrated seamlessly from Graviton3 to Graviton4 and now Graviton5, leveraging our Arm-based infrastructure investments with minimal friction at each step.

With Graviton5-based M9g instances, service latencies improved 27%, inter-core communication latency decreased 33%, and identical workloads achieved comparable throughput at just 52% CPU utilization versus 80% on previous generations. CockroachDB runs on Graviton across our customers' self-hosted fleet, and we look forward to broadly adopting M9g instances to deliver better price-performance, lower latencies, and greater efficiency for our customers worldwide."

Jeff White, Manager, Enterprise Architecture, Cockroach Labs

"After running AWS Graviton5 instances alongside our existing Graviton 4 fleet on CockroachDB gateway nodes, we saw a 23% reduction in CPU utilization and approximately 6% lower p99 latency — while handling 9% more queries per second. The performance gains were consistent across our entire evaluation window and scaled with load, which gives us confidence as we plan broader adoption."

Mick Giles, SVP, Technology, Hard Rock Bet

Honeycomb is an observability platform that provides real-time visibility and troubleshooting for cloud applications.

"Honeycomb achieved 36% better throughput per core on Graviton5 compared to Graviton4. We saw 28% lower latency and up to 26% less CPU utilization out of the box — and after optimizing CPU per task until latency matched across generations, we unlocked the full 36% throughput gain over 60 days of production ingest workloads." 

Liz Fong-Jones, Technical Fellow, Honeycomb.io

Learn more about Honeycomb.io migration experience.

"HubSpot has been running the majority of our workloads on Graviton instances since 2023. Our original motivation was cost — we saw approximately 15–20% reduction at comparable performance. Over time, we have increased Graviton adoption and the performance gains with every new instance type have been just as compelling. We recently deployed M9g instances in our pre-production environment and saw significant gains compared to M8g — especially for OLTP workloads on MySQL. For the databases in scope, query duration dropped by 45–60% alongside 20–30% lower CPU utilization. We look forward to adopting M9g instances at scale to further improve the performance our customers experience and drive down costs."

Philipp Hussels, Director, Foundational Infrastructure, HubSpot

"Joybuy is a European full-category online retail brand designed to bring customers a faster, more convenient, and cost-effective shopping experience. Our benchmarks show that AWS Graviton5-based M9g instances deliver significant performance gains over the previous Graviton4 generation across key database and caching workloads. In four typical Redis use cases, Graviton5 instances delivered an average 20% performance improvement. Across five representative MySQL workloads, Graviton5 achieved a 24% average performance uplift. For Elasticsearch, search latency was reduced by 37%. These results highlight Graviton5's generational leap in memory bandwidth and compute efficiency, making it a compelling upgrade for latency-sensitive, data-intensive workloads. Looking ahead, we plan to expand Graviton5 adoption in our current usage and explore its potential in additional data-intensive workloads."

Zhuangkai Zong, SRE Lead, Joybuy

KBR provides science, technology, and engineering solutions to governments and commercial clients worldwide.

"Graviton5 (M9g) delivered a 19% performance improvement over Graviton4 (M8g) on image processing workloads with zero code changes. Just a recompile with the latest ARM compiler flags, which unlocked significantly more auto-vectorization. Compared to Graviton3 (M7g), we're seeing roughly 30% faster scene processing."

Zachariah D. Dicus, Enterprise Solutions Architect, PMP, KBR

Match-Trade Technologies is a specialized B2B software developer that provides turnkey trading platforms and CRM solutions for financial markets.

"We benchmarked AWS Graviton5 (M9g instances) against Graviton4 (M8g instances) across our core trading workloads, and saw performance improvements ranging from 15% to 75%, with most workloads averaging 39–40% gains. These results exceeded our expectations and we look forward to expanding our Graviton5 footprint as it becomes available."

Krzysztof Macewicz, Head of DevOps, Match-Trade Technologies

“We’ve been working closely with AWS on running SAP HANA Cloud on AWS Graviton since 2023 and have seen significant performance improvements with each new Graviton generation. With AWS Graviton5-based Amazon EC2 M9g instances, we’ve observed a stunning 35-60% increase in the performance of our OLTP queries on SAP HANA Cloud - a phenomenal advancement in a single generation. We’re thrilled to be adopting the latest generation AWS Graviton instances for our SAP HANA Cloud deployment, as we expect this will deliver even greater performance gains.”

Stefan Bäuerle, Senior Vice President, Head of SAP HANA & Persistency

Calibre is Siemens EDA's industry-leading physical verification platform and the semiconductor industry's most trusted signoff solution. Serving as the final quality gatekeeper before manufacturing, Calibre performs critical DRC, LVS, and PERC verification, and advanced 3D IC multiphysics analysis to ensure chip designs meet foundry requirements at nanometer scales.

"The future of semiconductor physical verification lies in cloud-enabled, high-performance computing. Our deep collaboration with AWS positions Calibre at the forefront of this transformation, and we're excited to announce support for Calibre on Arm-based AWS Graviton processors. Today's deployments on AWS Graviton4 deliver 20% performance improvements and up to 30% compute cost reductions vs. other comparable AWS instances, and early AWS Graviton5 testing shows up to additional 30%+ performance boost, unlocking verification capabilities and faster time-to-market. Together, AWS and Siemens are building the infrastructure that will power the next decade of semiconductor innovation.”

Juan Rey, Senior Vice President and General Manager, Calibre Segment, Siemens EDA, Siemens Digital Industries Software

"SmugMug and Flickr are trusted worldwide to safely and securely store byte for byte copies of precious memories. Graviton5 provided up to a 20% speedup on CPU-bound image processing and compression workloads. The substantially larger CPU cache absorbs more of our working set, reducing main-memory traffic and allowing us to scale parallelism per instance. This results in lower operational cost while continuing to thrill our customers."

Michael Wallner, Senior Image Processing Engineer, SmugMug/Flickr

“Snowflake’s benchmarking of AWS Graviton5-based Amazon EC2 instances showed impressive performance gains across a wide range of workloads. Compared to prior generations, Graviton5 demonstrated notable improvements in compute efficiency, memory bandwidth, and overall throughput—delivering more than 30% performance boost on representative Snowflake workloads. These results highlight the continued evolution of AWS Graviton and its ability to power modern, data-intensive platforms efficiently. Graviton5’s strong balance of performance and energy efficiency aligns with Snowflake’s commitment to delivering high-performance, cost-effective, and sustainable data processing at scale. We’re excited to bring these advancements to our customers as we expand support for Graviton5 in future Snowflake releases.”

Neel Nadgir, Principal Engineer, Snowflake

“For over a decade since the inception of Annapurna Labs, Synopsys and AWS have collaborated to enable Amazon’s custom silicon development. Synopsys EDA tools such as VCS, PrimeTime, Fusion Compiler, and IC Validator support on AWS Graviton have been critical to the design of Graviton as well as Nitro and Trainium chips. Today, Synopsys and AWS are expanding Graviton support to accelerate our customers’ semiconductor innovation. Compared to Graviton4, early results on Graviton5 show up to 35% runtime improvements for Fusion Compiler and PrimeTime. In addition, our joint partner Arm is observing up to 40% faster runtimes for Synopsys VCS on Graviton5 relative to previous generations.”

Sanjay Bali, Senior Vice President, Strategy and Product Management, Synopsys

Tacnode Context Lake™ is an AI-native real-time data platform providing AI agents with millisecond-fresh context for real-time fraud/risk decisions and autonomous agents at scale.

"We benchmarked Graviton5-based M9g instances against our current Graviton4 (M8g) fleet using Tacnode's standard qualification workloads. Results are uniformly strong: throughput improved 20–30% across most workloads, and P99 tail latency on our most demanding context-serving paths more than halved under load — exactly the stability profile that high-stakes agentic decisioning demands. Graviton5's higher memory bandwidth is a particularly good match for Tacnode's bandwidth-sensitive mixed read/write paths at scale. Graviton5 will become the default compute tier for Tacnode on AWS."

Xiaowei Jiang, CEO & Chief Architect, Tacnode

Founded by the original creators of Apache Doris—one of the most popular open-source analytical databases—VeloDB delivers high-performance real-time analytics solutions to enterprises globally.

"We benchmarked Apache Doris on AWS Graviton5-based Amazon EC2 M9g instances against x86-based instances with identical configurations. The results were outstanding: Graviton5 delivered up to a 78% increase in query performance, with real-time analytics workloads improving by over 40% and wide-table aggregation exceeding 50%. These gains come from Doris's ARM-optimized vectorized engine taking full advantage of Graviton5's faster cores, larger caches, and higher memory bandwidth. For customers running analytical workloads at scale, the M9g offers a significant leap in both performance and cost-efficiency."

Liam, CEO, VeloDB