Meta Signs Multi-Billion-Dollar AWS Deal for Graviton Chips to Power Agentic AI

Meta has signed a multi-billion-dollar, multi-year agreement with Amazon Web Services to run a major share of its agentic AI workloads on AWS's custom Graviton processors. The deal, disclosed on April 24, will see Meta deploy hundreds of thousands of Graviton chips — equating to tens of millions of cores — over a period of at least three years.

Unlike most headline AI infrastructure deals, this one is built around CPUs rather than GPUs. Graviton is Amazon's Arm-based custom silicon, designed for the kind of CPU-intensive orchestration, retrieval and tool-use work that sits behind agentic AI systems. Meta said the chips will let it run those workloads with the performance and efficiency it needs at scale.

The agreement makes Meta one of AWS's top five Graviton customers and adds to a roughly $48 billion stack of recent AI infrastructure commitments Meta has made with CoreWeave and Nebius. It also signals that Meta's appetite for compute has outstripped the $135 billion capital expenditure budget it laid out for the year.

Amazon did not disclose an exact dollar figure, but multiple outlets pegged the contract at several billion dollars over its term. Investors took the announcement as validation of AWS's in-house silicon strategy: Amazon shares climbed roughly 3.5 percent in afternoon trading on April 26 as the news circulated and analysts upgraded the stock.

The deal is a notable diversification away from pure-Nvidia infrastructure for one of the world's largest AI buyers. By offloading agentic orchestration to Graviton, Meta can reserve scarce GPU capacity for training and large-scale inference, while AWS gets a flagship reference customer for selling Graviton as the CPU backbone of next-generation AI stacks.