Meta has committed to deploying tens of millions of AWS Graviton 5 CPU cores in a multi-year collaboration with Amazon Web Services, a deal that The Register reports would make the social network one of the largest-ever consumers of Amazon’s custom silicon. The compute will support Meta’s agentic AI deployments — specifically the software frameworks that run on top of GPU-accelerated generative AI models.
Santosh Janardhan, Meta’s head of infrastructure, characterized the deal as part of a deliberate compute diversification strategy. “As we scale the infrastructure behind Meta’s AI ambitions, diversifying our compute sources is a strategic imperative,” he said in a statement.
What Graviton 5 brings
Amazon’s Graviton 5 processors feature 192 of Arm’s Neoverse V3 cores, along with a substantially larger L3 cache and support for memory up to DDR5 8,800 MT/s. Amazon says that combination delivers a 25 percent performance uplift over Graviton 4.
The Register notes that while GPUs handle training and inference for generative AI models, the software frameworks that orchestrate those models still run on CPUs, making CPU capacity a distinct infrastructure concern as agentic workloads scale.
The Arm diversification pattern
The AWS deal fits into a broader shift in Meta’s infrastructure strategy that has been building for several months. According to The Register, Meta was among the first companies to deploy Nvidia’s standalone Grace CPUs at scale, announcing that in February. The company has also announced plans to deploy Nvidia’s newer 88-core Vera CPUs.
In March, Arm announced it had worked closely with Meta to design its first branded datacenter silicon — the “AGI CPU,” which packs 136 Neoverse V3 cores into a 300-watt part. That chip is not expected to arrive in Meta’s datacenters until later in 2026, according to The Register.
The broader Arm shift
Meta’s moves are part of a larger architectural transition in datacenter compute. Counterpoint Research analysts have predicted that by 2029, Arm-based CPUs will account for 90 percent of the AI ASIC server CPU market. Counterpoint analyst David Wu wrote that while x86 currently maintains a significant presence in AI server infrastructure, “our generation-by-generation analysis suggests this established stronghold is swiftly transitioning toward proprietary Arm-based designs.”
The Register points to several data points supporting that trajectory. Nvidia’s Grace CPUs, launched in 2023, have replaced x86 parts from Intel and AMD in many of Nvidia’s GPU systems. In December, AWS announced it was swapping out Intel CPUs in favor of its own silicon in its Trainium 3 AI rack systems. This week, Google said it would replace x86 chips in its TPU clusters with its own Arm-based Axion chips.
This article relies on a single source, The Register, which could not be independently corroborated.