AWS has opened general availability for its Graviton5 processor, offering customers a next-generation Arm-based chip designed specifically for agentic AI workloads. AWS states the chip provides up to 25% improved compute performance relative to the prior generation and features 192 cores per processor with 33% lower inter-core latency.
The Graviton5 is intended to support real-time reasoning, code generation and multi-step task orchestration. AWS has made EC2 instances powered by Graviton5 available: the M9g and M9gd families, which were first shown in a preview at re:Invent 2025 and are now accessible through standard EC2 adoption paths.
Several major customers have already begun deploying at scale. Meta has committed to using tens of millions of Graviton cores for its agentic AI efforts since the preview period. Other enterprise users including Uber and Snowflake are also deploying Graviton for agentic workloads, joining more than 120,000 customers who AWS says are building on the platform.
On measured workload comparisons provided by AWS, the M9g instances are positioned to deliver specific gains: 35% faster web applications, 35% faster machine learning inference and 30% faster databases versus the previous generation. The M9gd variant adds expanded local storage, offering up to 11.4 TB of NVMe SSD capacity along with up to 30% higher input/output operations per second than its predecessor.
Both instance types leverage the sixth-generation AWS Nitro System. Included in that platform is the Nitro Isolation Engine, which AWS describes as providing mathematically proven isolation between virtual machines. The Graviton5 platform supports DDR5-8800 memory speeds and PCIe Gen 6 connectivity.
For cloud customers and enterprise IT teams, the combination of higher core counts, reduced latency and increased local storage/I/O on the M9g and M9gd families represents a set of configuration choices meant to match a range of agentic AI and latency-sensitive workloads. The instances are now available through established EC2 channels for customers planning migrations or new deployments.
Technical details and availability
- Processor: 192 cores per Graviton5 processor
- Latency: 33% lower inter-core latency versus prior generation
- Performance: Up to 25% better compute performance for agentic AI workloads
- Memory and bus: Supports DDR5-8800 and PCIe Gen 6
- Instances: M9g (compute/memory balance) and M9gd (includes NVMe SSDs up to 11.4 TB)
- Platform: Sixth-generation AWS Nitro System with Nitro Isolation Engine
Adoption and customers
Meta, Uber and Snowflake are cited by AWS as early deployers, and the company reports that more than 120,000 customers are already building on the Graviton platform. M9g and M9gd instances move from preview to general availability and can be adopted through the standard EC2 provisioning paths.