Particle.news

Download on the App Store

Meta Partners With Arm to Shift AI Recommendations to Neoverse, Open-Source Optimizations

Meta targets lower energy use at hyperscale by standardizing key AI workloads on Arm CPUs.

Overview

  • Meta will migrate its ranking and recommendation systems that power personalization across Facebook and Instagram to Arm’s Neoverse data center platforms.
  • The companies say the Arm-based approach delivers higher performance with lower power consumption than x86 systems for these workloads.
  • Meta and Arm have adapted parts of Meta’s AI software stack for Arm architectures, including PyTorch and FBGEMM as well as ExecuTorch and vLLM, with improvements released as open source.
  • The agreement is a multi-year technical collaboration with no exchange of equity or physical infrastructure.
  • The move aligns with Meta’s expanding AI buildout, including a $1.5 billion Texas data center and multi‑gigawatt projects Prometheus and Hyperion, while Arm highlights a shift toward efficient cloud and on‑device inference as Arm CPUs gain share through deployments such as Nvidia’s Grace‑Blackwell systems.