Meta commits to tens of millions of AWS Graviton5 chips to cut AI infrastructure costs and reduce dependence on expensive GPUs

Yahoo Finance AI · April 25, 2026

AI Summary

•Meta signed a multi-year deal with Amazon Web Services to deploy tens of millions of Graviton5 processors — custom CPU chips that Amazon designed to run AI workloads more cheaply than traditional GPU-only setups. This marks Meta's biggest bet on CPU-heavy computing for its next-generation AI agents (self-deciding AI systems that can complete tasks without constant human prompts).
•Unlike GPU chips that excel at math-heavy calculations but cost thousands of dollars per unit, Graviton5 CPUs are cheaper to buy and run but require different software design. Meta is restructuring how it builds AI systems to work with this CPU-heavy mix, meaning some AI tasks will process slightly differently — but at a fraction of the power cost per request.
•For businesses using Meta's cloud services or AI APIs: cheaper infrastructure means Meta can lower prices or invest more in model quality without raising bills. For AWS: this locks in a major customer for years and proves its custom chips can compete in the AI arms race, weakening NVIDIA's near-monopoly on AI compute.

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.