Meta to Buy Tens of Millions of AWS Graviton5 Cores in Multi-Billion Dollar Deal
Meta will deploy tens of millions of AWS Graviton5 CPU cores, each containing 192 cores, in a multi-year agreement valued at billions of dollars. The chips will handle AI inference workloads post-training, signaling rising CPU demand and intensified competition in AI infrastructure.
1. Deal Overview
Meta has committed to use "tens of millions of cores" of AWS Graviton5 CPUs under a multi-year contract worth billions of dollars, marking one of the largest known non-GPU infrastructure deals in AI. This arrangement secures a steady supply of cost-effective CPU capacity for Meta’s expanding AI services.
2. Graviton5 Specs and Role
Each Graviton5 chip contains 192 cores that can be allocated to diverse workloads, with Amazon Web Services passing manufacturing savings on to customers. While GPUs remain essential for AI model training, these high-core-count CPUs will power inference tasks, reducing reliance on more expensive GPU instances.
3. Strategic Implications for AI Infrastructure
The deal underscores a broader AI-driven renaissance in the CPU market, with rising prices and demand for inference hardware. Meta’s move to diversify compute suppliers may temper future GPU requirements and reshape competitive dynamics, potentially influencing Nvidia’s inference revenue trajectory.