Meta deploys tens of millions of AWS Graviton CPU cores to power AI agents that reason and plan autonomously

Meta signed an agreement with AWS to deploy Graviton5 processors at scale, starting with tens of millions of cores. Graviton5 is a custom CPU chip (not GPU) built on 3-nanometer technology with 192 cores per chip and cache five times larger than the previous generation, enabling 25% better performance.

Unlike GPU-heavy model training, agentic AI workloads—where autonomous systems reason through multi-step tasks like code generation, search, and real-time problem-solving—are CPU-intensive. Graviton's larger cache cuts communication delays between cores by up to 33%, and its low-latency networking (Elastic Fabric Adapter) lets tasks distribute across many processors in coordination, making these workflows run faster and with better energy efficiency.

For businesses relying on Meta's services: Meta can now handle billions of AI-driven interactions more efficiently, reducing latency when AI agents execute complex multi-step tasks on your behalf. For infrastructure teams: this signals that scaling AI no longer means GPU-only strategies—purpose-built CPUs are becoming essential for the next wave of AI workloads, and diversifying compute sources (rather than relying on a single chip type) is now standard practice among hyperscalers.