Amazon Web Services (AWS) has made its Graviton5-powered Amazon EC2 C9g and C9gd instances generally available, bringing its latest Arm-based processor to compute-optimized instances aimed at AI inference, high-performance computing, distributed analytics, and other CPU-intensive workloads.
The company says C9g delivers up to 25% higher performance per vCPU than the previous-generation C8g family, along with DDR5-8800 memory support, five times the L3 cache, PCIe Gen 6 connectivity, and up to three times higher packet-processing performance.
AWS is also rolling out platform-level updates via its latest Nitro platform, including Nitro Isolation Engine, a Rust-based security capability in the Nitro Hypervisor that mediates access to memory, CPU register state, and I/O devices to strengthen VM isolation.
Why It Matters for Operators
The launch reflects more than a routine processor refresh. While GPUs remain the primary engines for model training and large-scale inference, the rest of the AI stack is evolving into agentic systems that plan, call tools, and execute multi-step workflows. This places more responsibility on CPUs for orchestration, memory management, scheduling, and concurrency, requiring operators to balance compute, memory, networking, and storage so expensive accelerators remain fully utilized.
AWS is explicitly positioning C9g/C9gd for that role, and Industry analysts say the company is reflecting a broader architectural trend.
“AI enablement is a design point for every CPU moving forward,” Matt Kimball, vice president and principal analyst for data center technologies at Moor Insights & Strategy, told Data Center Knowledge. “While all of the hype is on GPUs, CPU demand is going to continue to grow at a very rapid rate.”
Kimball noted that Graviton5 is designed to serve both established enterprise workloads and emerging AI applications. Larger caches, faster memory, and higher-bandwidth I/O benefit databases, analytics, and HPC, while the processor’s architecture also suits CPU-bound AI tasks such as reasoning, task decomposition, and concurrency.
AWS Pushes Deeper Into Infrastructure
The launch builds on AWS’s broader push to expand Graviton’s role across its cloud platform. Earlier this year, the company introduced Graviton-powered Amazon Redshift RG instances to improve the performance and economics of AI-era analytics. With the C9g family, AWS is pushing that strategy deeper into the infrastructure layer, optimizing CPUs for the orchestration and memory-heavy paths that surround AI inference.
Stephen Sopko, semiconductor and deep tech analyst at HyperFrame Research, said the shift reflects the changing nature of AI workloads rather than any diminished role for accelerators.
“Model inference may run on accelerators, but orchestration, tool calling, and multi-step reasoning are CPU-bound,” Sopko said. “It doesn’t replace accelerators; it keeps them fed while agents plan, call tools, and hold state.”
Sopko expects that balance to matter more as enterprises deploy larger agentic AI systems. “Agentic AI turns the CPU from supporting actor back into co-star,” he said. “C9g is Amazon wiring the control plane for that shift.”
Competition Expands Beyond x86
AWS is not alone in redesigning server processors around AI. Nvidia’s Grace CPU is tightly integrated with the company’s AI platform, Arm has emphasized AI throughout its Neoverse roadmap, and Qualcomm has introduced server CPU variants for orchestration, general-purpose computing, and AI head-node functions.
Kimball said Graviton5’s closest competitors are still AMD EPYC and Intel Xeon families rather than AI accelerators. Even so, he expects every major CPU line to continue evolving around AI workloads as demand grows for processors optimized to manage data movement, memory, and orchestration alongside GPUs.
C9g/C9gd support PCIe Gen 6 and the latest Nitro platform, with instances offered in 11 sizes from medium through 48xlarge, including bare-metal options. According to AWS, the largest instances provide up to 100 Gbps of networking and up to 72 Gbps of Amazon EBS bandwidth. C9gd instances also include local NVMe SSD storage for workloads that require high-speed caching or scratch space.
The instances are initially available in the AWS US East (Northern Virginia and Ohio), US West (Oregon) and Europe (Frankfurt) regions, with additional regions planned.
