Meta deploys hundreds of Graviton chips

Meta signed a new agreement with Amazon Web Services to run agentic artificial intelligence workloads on AWS Graviton processors. (aboutamazon.com) Amazon said the deployment begins with tens of millions of Graviton cores and can expand as Meta adds more AI capacity. CNBC reported the commitment covers hundreds of thousands of chips over at least three years. (aboutamazon.com) (cnbc.com) The chips are meant for “agentic” AI work, which uses software systems that reason through steps, call tools, and coordinate tasks in real time. Amazon said those jobs create heavy demand for central processors, not just graphics processors used to train large models. (aboutamazon.com) (geekwire.com) Amazon said Meta is now one of the largest Graviton customers in the world. The companies said the agreement builds on Meta’s existing use of Amazon Bedrock and a longer AWS relationship. (aboutamazon.com) Graviton is Amazon’s in-house central processor family for cloud servers, built to lower cost and power use on workloads that do not need a graphics chip for every step. AWS says more than 90,000 customers use Graviton-based services. (aws.amazon.com) (aboutamazon.com) The timing lines up with Amazon’s broader push to show its custom chip business is no side project. Chief executive Andy Jassy said in his April 9 shareholder letter cycle that Amazon’s custom silicon business is running at more than $20 billion in annual revenue. (cnbc.com) (ir.aboutamazon.com) Meta has been signing multiple infrastructure deals as it expands AI capacity. CNBC said the Graviton agreement followed roughly $48 billion in recent AI infrastructure commitments with CoreWeave and Nebius. (cnbc.com) GeekWire, citing Bloomberg, described the Graviton agreement as a multibillion-dollar deal. Meta infrastructure chief Santosh Janardhan said diversifying compute sources is a “strategic imperative” as the company scales its AI systems. (geekwire.com) The deal puts Amazon deeper inside Meta’s AI stack without replacing the graphics processors still used for model training. It gives Amazon a marquee outside customer for Graviton as both companies spend heavily to build the next layer of AI infrastructure. (aboutamazon.com) (geekwire.com)