Amazon’s AI chip push gives rivals a new problem

Amazon Web Services (AWS) is expanding its custom silicon strategy with the broad release of Graviton5-powered EC2 instances, designed to meet the demands of agentic AI. These chips, featuring 192 cores per CPU and improved memory efficiency, target workloads requiring real-time reasoning, code generation, and multi-step task coordination. AWS claims the new M9g instances deliver up to 25% better computing performance than the previous generation, with benefits extending to web apps, machine learning inference, and database speeds. The M9gd instances, optimized for high-speed local storage, offer up to 11.4 terabytes of NVMe SSD storage and 30% more input/output operations per second. This aligns with AWS’s broader effort to provide cost-effective AI infrastructure, as enterprises seek to balance performance and expenses amid rising demand. Unlike competitors relying on external chip providers, AWS designs its own silicon, allowing deeper integration with its cloud stack. Graviton5 complements AWS’s existing Trainium chips for AI training and Nvidia GPUs for acceleration, reinforcing its position in the AI hardware race. The launch underscores a shift in cloud competition, where underlying hardware—particularly CPUs—is becoming as critical as software and storage. AWS’s move reflects the growing importance of agentic AI, which demands faster orchestration and real-time processing beyond traditional model training. With AI infrastructure becoming a key battleground, AWS’s Graviton5 aims to address enterprise needs for scalable, high-performance solutions without spiraling costs. The chip’s focus on efficiency and multi-core communication positions it as a tool for next-generation AI workloads.