AWS Graviton5 Delivers 35% AI Speed Boost, Revolutionizing Cloud Computing

AWS's Graviton5 custom silicon redefines cloud price-performance, significantly boosting AI and general workloads for unprecedented efficiency.

December 5, 2025

AWS Graviton5 Delivers 35% AI Speed Boost, Revolutionizing Cloud Computing
Amazon Web Services has escalated the competitive landscape of cloud computing with the announcement of Graviton5, its most powerful and efficient custom-designed CPU to date. Unveiled at the company's re:Invent conference, the new processor promises significant performance and efficiency gains for a wide array of workloads running on its Elastic Compute Cloud (EC2) platform.[1][2] The Graviton5 will initially power new Amazon EC2 M9g instances, which are now available in a preview phase.[3][4] This launch marks a continued strategic push by AWS into developing its own silicon, aiming to provide customers with better price-performance and further optimize its vast cloud infrastructure. For the third consecutive year, custom Graviton chips have accounted for more than half of all new CPU capacity added to AWS.[5][6][7]
The technical advancements embodied in the Graviton5 are substantial, representing a significant leap over its predecessor, the Graviton4. Built on advanced 3nm process technology, the new chip boasts a remarkable 192 cores in a single package, the highest core density available in Amazon EC2.[8][5][9] This dense, single-socket design is a key architectural choice, reducing the data travel distance between cores and cutting inter-core communication latency by up to 33%.[8][4][10] A major highlight is the five-fold increase in the size of the L3 cache, a critical high-speed memory buffer.[11][1][6] This expansion means each core has access to 2.6 times more L3 cache than in the Graviton4, leading to faster data access and improved application responsiveness.[8][12][13] Furthermore, Graviton5 supports faster memory speeds and offers increased bandwidth for both networking and storage, with up to 15% higher network bandwidth and 20% higher Amazon Elastic Block Store (EBS) bandwidth on average.[8][9][11] For the largest instance sizes, the network bandwidth can even double.[8][1][7]
These hardware enhancements translate into significant real-world performance improvements across a variety of applications. AWS claims that the new M9g instances, powered by Graviton5, deliver up to 25% better overall compute performance compared to the previous generation M8g instances based on Graviton4.[3][14][12][6] The company has released specific metrics highlighting these gains, noting that web applications could see a speed increase of up to 35%, databases could run up to 30% faster, and crucially for the AI sector, machine learning workloads can experience a performance boost of up to 35%.[3][15][14] Early results from major AWS customers corroborate these claims. Atlassian, for instance, observed 30% higher performance and 20% lower latency for its Jira software on M9g instances.[14][8] Similarly, SAP reported a stunning 35% to 60% performance increase for its OLTP queries on SAP HANA Cloud.[14][8] Other industry leaders like Adobe, Airbnb, Epic Games, and Pinterest are already leveraging Graviton chips for demanding tasks, from real-time gaming to large-scale data analytics.[12][6]
The introduction of Graviton5 carries profound implications for the artificial intelligence industry. While specialized chips like GPUs and AWS's own Trainium processors are essential for training large AI models, general-purpose CPUs like Graviton play a critical role in the broader AI workflow, particularly for inference and data processing. The enhanced performance and efficiency of Graviton5 make it an attractive option for running CPU-based machine learning inference, especially for smaller language models and other AI applications where cost-effectiveness is paramount.[16][17] The significant memory bandwidth and cache improvements are particularly beneficial for memory-intensive AI workloads. The processor's power efficiency, a result of the 3nm technology and system-level optimizations like bare-die cooling, also allows companies to scale their AI operations while managing costs and adhering to sustainability goals.[8][9][13] The new instances are built on the AWS Nitro System, which offloads virtualization, storage, and networking functions to dedicated hardware, and introduces a "Nitro Isolation Engine" that uses formal verification to provide mathematical certainty of workload isolation, enhancing security for sensitive AI data.[8][13][1]
In conclusion, the launch of AWS Graviton5 and the accompanying M9g instances marks a significant milestone in the evolution of cloud computing hardware. By continuing to push the boundaries of custom silicon design, AWS is not just offering incremental upgrades but is fundamentally altering the price-performance equation for its customers. The combination of higher core counts, massive cache improvements, and increased bandwidth delivers tangible benefits for a wide range of general-purpose workloads. For the AI industry, Graviton5 provides a powerful and efficient platform for inference and data-intensive tasks, enabling broader adoption and more cost-effective scaling of AI applications. As AWS expands its Graviton5-powered offerings with compute-intensive (C9g) and memory-intensive (R9g) instances planned for 2026, the impact of this custom CPU on the cloud and AI landscape is set to grow even further.[12][11][1]

Sources
Share this article