ARM Server Architecture

Overview

ARM Server Architecture refers to the use of ARM instruction-set CPUs in server and rack-scale data center platforms, in contrast to the dominant x86 architecture (Intel Xeon, AMD EPYC). Key current implementations include Ampere Altra/AltaMax (tailored for cloud-native and web-tier workloads), NVIDIA Grace CPU (paired with Blackwell GPUs in the NVL72/GB200 platforms for AI training and inference), and the forthcoming NVIDIA Vera CPU (announced at Computex 2026 for agentic AI workloads).

As of 2025, ARM adoption in data centers is an active but not yet mainstream transition. Hyperscalers and infrastructure vendors are evaluating ARM CPUs for specific workload categories — particularly workloads that favor high core counts and power efficiency over single-thread performance. Manufacturers like Jabil are actively polling customers on ARM journey maturity (Ampere vs. NVIDIA Grace/Vera vs. staying on x86) to determine building-block roadmap priorities. The OCP MHS standard already includes Ampere and SP7 (ARM) as supported HPM options alongside Intel and AMD platforms.