Meta picks AWS Graviton CPUs
- Meta signed an agreement on April 24 to deploy Amazon Web Services Graviton processors at scale for its next generation of artificial intelligence systems, expanding a long-standing cloud partnership with Amazon. - Amazon said the rollout starts with tens of millions of Graviton cores and could grow further, making Meta one of the largest Graviton customers as it builds agentic AI workloads. - The deal shifts attention toward CPUs for inference and orchestration after years of GPU-heavy spending on AI training. (aboutamazon.com)
Meta signed a new agreement to run parts of its next-generation artificial intelligence stack on Amazon Web Services’ Graviton processors. (aboutamazon.com) Amazon said on April 24 that the deployment starts with tens of millions of Graviton cores, with room to expand as Meta’s AI use grows. (aboutamazon.com) The companies said the chips will support “agentic” workloads, the software that chains together multiple steps such as search, coding, and real-time reasoning after a model has already been trained. (aboutamazon.com) (aws.amazon.com) That makes this a different part of the AI stack from model training, where graphics processing units, or GPUs, still dominate. Amazon’s own guidance separates training from inference, the stage when a model is serving live requests. (aws.amazon.com) TechCrunch reported that Meta’s commitment covers millions of AWS Graviton chips and framed the move as a bet that more AI work will be handled by central processing units, or CPUs, once models are in production. (techcrunch.com) Amazon said the agreement builds on Meta’s existing use of Amazon Bedrock at scale, alongside a longer AWS relationship that predates the new Graviton rollout. (aboutamazon.com) Graviton is Amazon’s Arm-based server chip family, built for cloud workloads rather than sold as stand-alone hardware. Amazon said its newest Graviton5 chip has 192 cores and that more than half of new CPU capacity added to AWS has been powered by Graviton for three straight years. (aboutamazon.com 1) (aboutamazon.com 2) Amazon has been using similar cloud-and-chip agreements to lock in AI customers. TechCrunch reported this week that Anthropic agreed to spend more than $100 billion on AWS over 10 years, with a focus on Amazon’s Trainium accelerators. (techcrunch.com) The Meta agreement gives Amazon a marquee customer for Graviton just as cloud providers and chipmakers push a broader message: AI data centers need CPUs for routing, serving, and coordinating model calls, not only GPUs for training. (techcrunch.com) (aws.amazon.com) For Meta, the deal adds another large cloud supplier to an AI infrastructure strategy that already spans multiple providers. TechCrunch noted that Meta signed a six-year, $10 billion Google Cloud agreement last August. (techcrunch.com) The immediate result is simple: more of Meta’s live AI traffic is headed to Amazon’s homegrown CPUs, not just the industry’s scarce and expensive GPUs. (aboutamazon.com) (techcrunch.com)
