Meta's Graviton Push
- Meta said Friday, April 24, that it signed an agreement with Amazon Web Services to deploy AWS Graviton processors at scale for agentic artificial intelligence workloads across Meta’s services. - The rollout starts with tens of millions of Graviton cores and can expand further, making Meta one of the world’s largest Graviton customers as it broadens beyond graphics processors. - The deal extends Meta’s multi-chip strategy after recent custom-silicon moves with Broadcom and MTIA, spreading AI work across CPUs, GPUs and in-house chips. (about.fb.com)
Meta said April 24 that it will add AWS Graviton processors to its compute fleet to run agentic artificial intelligence workloads. (about.fb.com) (press.aboutamazon.com) The deployment begins with tens of millions of Graviton cores, with room to expand as Meta’s AI capacity grows. Amazon said that makes Meta one of the largest Graviton customers in the world. (press.aboutamazon.com) Graviton is Amazon’s Arm-based central processor, the kind of chip that handles general-purpose computing rather than the matrix-heavy work usually pushed onto graphics processors. Network World reported the deal centers on Graviton5, with 192 cores per chip. (press.aboutamazon.com) (networkworld.com) Amazon and Meta said those CPUs will be used for agentic AI jobs such as real-time reasoning, code generation, search and coordinating multi-step tasks. Those workloads lean on orchestration and memory management, not just raw model training. (press.aboutamazon.com) (networkworld.com) That makes this a compute-mix story as much as a cloud deal. Meta has been saying for weeks that no single chip architecture can serve every artificial intelligence workload efficiently. (about.fb.com 1) (about.fb.com 2) On March 11, Meta said it was developing and deploying four new generations of its Meta Training and Inference Accelerator chips within two years. It said those chips would support ranking, recommendations and generative AI workloads, with MTIA 300 already in production. (about.fb.com) On April 14, Meta said Broadcom would help co-develop multiple generations of next-generation MTIA chips and support a first deployment exceeding 1 gigawatt. Meta said that rollout would grow to multiple gigawatts over time. (about.fb.com) The AWS agreement adds another lane to that buildout: cloud CPUs for the control-plane work around large AI systems, while custom accelerators and graphics processors handle other jobs. Network World described Meta’s stack as spanning AWS, Nvidia, AMD, Arm and Meta’s own silicon. (networkworld.com) (about.fb.com) Meta framed the AWS deal as an extension of a long-standing relationship and said it already uses Amazon Bedrock at scale. The immediate result is more CPU capacity for the parts of AI systems that have to coordinate billions of interactions, not just generate tokens. (press.aboutamazon.com)
