Meta partners up with Arm to scale AI efforts (techcrunch.com)

Meta announced a multi-year partnership with Arm to move its ranking and recommendation systems onto Arm’s Neoverse platform, which Arm recently optimized for cloud AI workloads. The deal aims to help Meta scale AI across its apps for more than 3 billion users without exchanging ownership stakes or handing over physical infrastructure—unlike some recent vendor-funded compute deals. The timing dovetails with Meta’s massive data-center expansion (projects “Prometheus” and “Hyperion”), which target multiple gigawatts of power delivery by 2027–2030 and include a 200 MW natural-gas plant to directly serve capacity under construction in Ohio. Technically, the shift signals stronger interest in energy-efficient, Arm-based inference and distributed AI compute rather than exclusive reliance on high-power accelerators. Arm is pitching Neoverse’s performance-per-watt and low-power advantages for large-scale recommendation and ranking workloads that are latency- and throughput-sensitive across many servers. For the AI ecosystem, the move underscores a maturing, heterogeneous hardware stack—Nvidia, AMD and Arm will all compete to supply cloud and hyperscaler AI needs—affecting procurement, power planning, and software optimization (compilers, runtimes, model quantization and operator kernels) for inference at enormous scale.