AWS launches EC2 C9g instances with Graviton5 processors, 25% faster

What happened?

On May 6, 2025, AWS launched EC2 C9g and C9gd instances, the first based on its own Graviton5 processors. These instances are optimized for compute-intensive workloads such as real-time analytics, batch processing, video encoding, scientific modeling, and CPU-based machine learning inference. This launch marks a milestone in AWS's strategy to reduce dependence on Intel and AMD, continuing the evolution that began with Graviton in 2018. Since then, each generation has offered substantial improvements: Graviton2 (2020) introduced 40% more performance than equivalent x86 instances, Graviton3 (2022) added bfloat16 support and improved ML performance, and Graviton4 (2023) doubled memory bandwidth. With Graviton5, AWS aims to consolidate its leadership in efficient computing for the age of agentic AI.

Key specifications

C9g instances offer up to 25% more performance per vCPU compared to C8g (Graviton4). They feature DDR5 memory at 8800 MT/s, the fastest of any cloud instance, 5 times more L3 cache, and up to 3 times better packet processing performance. They are available in 11 sizes, from medium to 48xlarge, plus a bare metal option. The C9gd variant adds high-speed local NVMe SSD storage. In detail, network bandwidth reaches up to 100 Gbps (versus 50 Gbps on C8g) and EBS bandwidth up to 72 Gbps (versus 40 Gbps). DDR5 memory at 8800 MT/s represents a leap from 4800 MT/s on Graviton4, reducing data access latency in applications like in-memory analytics and real-time databases. Additionally, C9gd instances include detailed NVMe metrics with latency histograms by I/O size and 1-second granularity, accessible via CloudWatch, enabling optimization of latency-sensitive workloads.

Why is this important?

This is a significant generational leap. Faster memory and larger caches reduce data access latency, improving performance in applications like in-memory analytics and AI agent loops. Increased network bandwidth (up to 100 Gbps) and EBS bandwidth (up to 72 Gbps) eliminate bottlenecks in distributed workloads. According to AWS, C9g instances are ideal for agentic AI workloads, where CPU is critical for reasoning and task orchestration. As AI moves from answering questions to executing actions and coordinating multi-step tasks, CPU compute demand grows, and these instances are designed for that shift. Moreover, detailed NVMe metrics support allows developers to fine-tune I/O performance for applications such as HPC simulations, ML inference caches, or local buffers in ad engines.

Market implications

With Graviton5, AWS reinforces its commitment to custom silicon, directly competing with Intel and AMD. For customers, this translates to more efficient compute options and potentially lower cost per performance. Historically, Graviton instances have offered up to 40% better price-performance than equivalent x86 instances. With Graviton5, that gap could widen, pressuring traditional chipmakers to innovate faster. Additionally, the focus on agentic AI positions AWS to capture a growing share of ML inference spending, which Gartner predicts will account for 30% of total cloud spending by 2026. However, AWS's reliance on its own silicon also carries vendor lock-in risks, though ARM compatibility and the open-source ecosystem partially mitigate this.

What readers should know

The instances are now available in US East (N. Virginia), US West (Oregon), and Europe (Ireland). They can be launched under on-demand, reserved, or spot models. AWS has also improved NVMe metrics, offering detailed performance statistics via CloudWatch. Pricing has not been officially announced, but is expected to follow the trend of previous generations: similar or slightly lower cost per vCPU than C8g, with 25% more performance. For workloads requiring local storage, C9gd instances offer high-speed NVMe SSDs with detailed metrics. It is recommended to evaluate C9g for applications such as batch processing, video encoding, distributed analytics, and CPU-based ML inference. For those migrating from C8g, AWS provides migration guides and tools like AWS Graviton Ready to validate compatibility.