Meta inks large-scale deal to run AI workloads on AWS Graviton CPUs
- Meta said on April 24 it signed a multiyear agreement with Amazon Web Services to deploy Graviton processors at scale for agentic AI workloads. - The rollout starts with tens of millions of Graviton cores, equal to hundreds of thousands of chips, in a deal CNBC said lasts at least three years. - The pact adds CPUs to Meta’s recent GPU buying spree and broadens its AI compute mix. (cnbc.com)
Meta said on April 24 that it will use Amazon Web Services’ Graviton processors at scale to run agentic artificial intelligence workloads. (about.fb.com) (aboutamazon.com) The deployment starts with tens of millions of Graviton cores and can expand as Meta’s AI demand grows, according to Meta and AWS. CNBC reported the agreement runs for at least three years. (about.fb.com) (cnbc.com) AWS said the deal makes Meta one of the largest Graviton customers in the world. CNBC reported that means hundreds of thousands of chips, placing Meta among AWS’s top five Graviton customers. (aboutamazon.com) (cnbc.com) A central processing unit, or CPU, handles many small jobs at once, while a graphics processing unit, or GPU, is built for the heavy math used to train large AI models. Meta and AWS said newer agent systems also need large amounts of CPU capacity for search, planning, code generation, and multi-step task orchestration. (aboutamazon.com) (cnbc.com) AWS is pitching Graviton5 for that layer of work. Amazon said the chip has 192 cores and a cache five times larger than the prior generation, cutting some communication delays by up to 33%. (aboutamazon.com) Meta’s infrastructure push has been widening beyond one supplier. CNBC said the Graviton agreement follows about $48 billion in recent AI infrastructure commitments with CoreWeave and Nebius, both tied to Nvidia graphics processors. (cnbc.com) AWS also tied the new deal to Meta’s existing use of Amazon Bedrock, its managed AI service, and to a longer-running cloud relationship between the companies. That gives Amazon a marquee customer for its in-house Arm-based silicon as cloud providers try to prove they can sell more than Nvidia access. (aboutamazon.com) (about.fb.com) The immediate result is not that Meta is replacing GPUs. It is adding a large CPU pool for the parts of AI systems that sit before, after, and around model training, where cost and energy use can matter as much as raw speed. (aboutamazon.com) (cnbc.com)
