Meta signs Graviton5 chip deal

Meta said on April 24 that it will add Amazon Web Services Graviton processors to its artificial intelligence infrastructure for agentic AI workloads. (about.fb.com) (press.aboutamazon.com) The agreement starts with tens of millions of Graviton cores, with room to expand as Meta’s AI needs grow. Amazon said the deployment will make Meta one of the largest Graviton customers in the world. (press.aboutamazon.com) CNBC reported the deal runs for at least three years and covers hundreds of thousands of chips. AWS Vice President Nafea Bshara told CNBC the move puts Meta among the top five Graviton customers. (cnbc.com) Graviton chips are central processing units, or CPUs, the general-purpose processors that handle scheduling, memory, and other system work. Meta and AWS are pitching them for AI jobs that involve real-time reasoning, search, code generation, and coordinating multi-step tasks. (press.aboutamazon.com) (networkworld.com) That is a different role from Nvidia-style graphics processing units, or GPUs, which still do most of the heavy lifting for training large models. CNBC reported AWS also sees Graviton as useful for post-training and other AI refinements after the biggest training runs are done. (cnbc.com) AWS launched Graviton5 in December, saying each chip has 192 cores and delivers up to 25% higher performance than the prior generation. AWS also said Graviton5-based systems can improve price-performance while using less energy than alternatives. (aboutamazon.com) (aws.amazon.com) For Meta, the AWS pact adds another supplier to a fast-growing compute portfolio. Network World reported the company is already working across Nvidia, Advanced Micro Devices, Arm, and its own Meta Training and Inference Accelerator chips. (networkworld.com) The timing lines up with a much larger spending push. Meta said in January that it expects 2026 capital expenditures of $115 billion to $135 billion, up from $72.22 billion in 2025, as it builds more AI infrastructure. (datacenterdynamics.com) Amazon is also using the deal to show that its homegrown silicon can win major external customers, not just power Amazon’s own cloud. CNBC reported Graviton already counts Adobe, Apple, and Snowflake among users, and AWS said earlier this week that Anthropic plans to use Graviton processors too. (cnbc.com) The immediate result is that Meta is buying more CPU capacity, not replacing GPUs. The larger signal is that big AI buyers are spreading workloads across different kinds of chips instead of relying on one architecture for everything. (networkworld.com)