Amazon Web Services has moved its Graviton5-powered EC2 C9g and C9gd instances into general availability, giving customers access to the company’s latest Arm-based processor in a compute-optimized form factor designed for AI inference, high-performance computing, distributed analytics, and other CPU-intensive tasks.
What Graviton5 Brings to the Table
AWS says the C9g family delivers up to 25% higher performance per vCPU compared to the previous C8g generation. The processor pairs DDR5-8800 memory support with five times the L3 cache of its predecessor, PCIe Gen 6 connectivity, and up to three times higher packet-processing throughput. The largest instance sizes offer up to 100 Gbps of networking and up to 72 Gbps of Amazon EBS bandwidth. C9gd instances add local NVMe SSD storage for workloads that need high-speed caching or scratch space.
The platform also introduces the Nitro Isolation Engine, a Rust-based security capability built into the Nitro Hypervisor. It mediates access to memory, CPU register state, and I/O devices with the goal of strengthening virtual machine isolation at the hardware level.
Instances are available in 11 sizes, from medium through 48xlarge, including bare-metal options. Initial availability covers US East (Northern Virginia and Ohio), US West (Oregon), and Europe (Frankfurt), with additional regions planned.
The CPU’s Growing Role in AI Infrastructure
The launch is not simply a generational refresh. AWS is positioning C9g and C9gd specifically for the orchestration and control-plane work that surrounds GPU-based inference. As AI systems evolve toward agentic architectures that plan multi-step workflows, call external tools, and manage state across long-running tasks, CPUs absorb a larger share of scheduling, memory management, and concurrency responsibilities.
Matt Kimball, vice president and principal analyst for data center technologies at Moor Insights and Strategy, described AI enablement as a core design point for every CPU going forward. He told Data Center Knowledge that CPU demand will continue to grow at a rapid rate even as GPU spending dominates the headlines, and that Graviton5 is built to serve both established enterprise workloads and emerging AI applications. Larger caches, faster memory, and higher-bandwidth I/O benefit databases and HPC, while the architecture also suits CPU-bound AI tasks such as reasoning and task decomposition.
Stephen Sopko, semiconductor and deep tech analyst at HyperFrame Research, framed the dynamic clearly: model inference may run on accelerators, but orchestration, tool calling, and multi-step reasoning are CPU-bound work. The CPU’s job, in his view, is to keep accelerators fed while agents plan and hold state, not to replace them.
A Competitive Landscape Beyond x86
AWS is not the only vendor rethinking server CPU design around AI. Nvidia’s Grace CPU is tightly coupled with its accelerator platform, Arm has emphasized AI throughout its Neoverse roadmap, and Qualcomm has introduced server CPU variants targeting orchestration and AI head-node functions. Kimball noted that Graviton5’s most direct competition remains AMD EPYC and Intel Xeon families rather than GPU accelerators, though he expects every major CPU line to continue evolving to handle data movement, memory, and orchestration workloads alongside GPUs.
The C9g launch extends a broader Graviton push at AWS that earlier this year also brought Graviton-powered Amazon Redshift RG instances to analytics workloads. Together, the moves signal that AWS sees its custom Arm processors as infrastructure for the full AI stack, not just general-purpose compute.