Meta buys Graviton cores

Meta said Friday, April 24, that it will deploy tens of millions of Amazon Web Services Graviton cores for the CPU-heavy work behind its agentic artificial intelligence systems. (about.fb.com) (press.aboutamazon.com) The companies said the first rollout will use AWS Graviton5, Amazon’s Arm-based server chip, and will make Meta one of the largest Graviton customers in the world. (about.fb.com) (press.aboutamazon.com) A core is one processing unit inside a central processor, the general-purpose chip that handles tasks like scheduling, search, and code execution. Meta said those jobs are becoming more important as its artificial intelligence agents take on multi-step reasoning and task orchestration. (about.fb.com) (press.aboutamazon.com) Meta and Amazon both framed the agreement as a response to a shift inside artificial intelligence computing: graphics processors still dominate model training, but agent systems also need large fleets of CPUs. Amazon said those workloads include real-time reasoning, code generation, search, and coordinating multi-step tasks. (press.aboutamazon.com) Meta infrastructure chief Santosh Janardhan said the company is diversifying compute sources as a “strategic imperative,” and said Graviton will help run CPU-intensive artificial intelligence workloads at Meta’s scale. AWS vice president Nafea Bshara said the partnership combines custom silicon with Amazon’s broader artificial intelligence stack. (about.fb.com) The agreement also fits Meta’s wider chip strategy. On March 11, Meta said it was developing four new generations of its Meta Training and Inference Accelerator chips over the next two years, with MTIA 300 already in production and later versions aimed at generative artificial intelligence inference through 2027. (about.fb.com) In that March update, Meta said it deploys hundreds of thousands of MTIA chips for inference across ads and organic content, while also sourcing silicon from outside partners. The AWS deal adds another supplier rather than replacing Meta’s in-house hardware program. (about.fb.com 1) (about.fb.com 2) For Amazon, the announcement is a showcase for Graviton as more than a cost-saving cloud processor. AWS said it now offers more than 150 Graviton-powered EC2 instance types, has built more than 2 million Graviton processors, and serves more than 50,000 Graviton customers. (aws.amazon.com) AWS has said Graviton4 delivers up to 30% better performance than comparable Graviton3-based R7g instances on one memory-optimized server family. Meta’s new deal suggests Amazon is now pitching that Arm-based line into the center of big-company artificial intelligence operations, not just conventional cloud workloads. (aws.amazon.com) (press.aboutamazon.com) The immediate result is simple: Meta is buying a very large pool of general-purpose compute from Amazon to handle the non-GPU work its artificial intelligence products increasingly require. The companies left the financial terms undisclosed and said the deployment can expand from its initial tens of millions of cores. (about.fb.com) (press.aboutamazon.com)