The Graviton5 processor is designed to handle CPU-intensive AI workloads like code generation, real-time reasoning, search, and coordinating multi-step tasks, signaling a shift from GPUs for training large models to advanced CPU resources.

Driven by agentic AI workloads that require CPU-heavy processing for real-time reasoning and orchestration of multi-step tasks beyond what GPUs can provide, Graviton5 targets CPU-centric AI scenarios.

Optimized for CPU-heavy workloads and multi-step task execution, Graviton5 addresses the rising demand for CPU power in agentic AI environments.

The agreement covers the entire infrastructure stack, including infrastructure, networking, power, and management layers for hosted Graviton cores, not just the chips.

Built on 3‑nanometer technology, Graviton5 delivers up to 25% infrastructure performance improvements over the prior generation and enhances energy efficiency.

The system supports Elastic Fabric Adapter (EFA) to enable low‑latency, high‑capacity connections between multiple virtual environments for coordinating broad agentic AI workloads across processors.

Graviton5 operates within the AWS Nitro system to maximize performance, availability, and security, with direct hardware access and features like ENA and EBS for private virtual environments.

Meta and AWS have signed an agreement to deploy tens of millions of Graviton processors to support Meta’s next‑generation AI initiatives.

Meta commits to tens of millions of AWS Graviton Arm cores as part of a wholesale infrastructure deal, signaling a major expansion of the Meta–AWS partnership.

Graviton5 is designed for sustained processing with high bandwidth, featuring 192 Arm Neoverse V3 cores, 600 MB cache, and support for DDR5-8800 memory and PCIe Gen6.

Graviton5 provides a cache five times larger than the previous generation, reducing core response time by up to 33% and enabling faster data processing with greater bandwidth.

Santosh Janardhan, Meta’s head of infrastructure, frames the move as a strategic step to diversify compute sources to support Meta’s AI scaling ambitions.