Meta’s Muse AI agent reached about 2 million daily active users by Oct. 5, shortly after its launch, and ranked among Apple’s most-downloaded apps.
The early adoption has focused attention on the hardware needed to run AI agents. GPUs remain important for training models and handling inference, but agents also require sequential logic for planning, retrieving information, calling tools and coordinating services. That could increase the role of server CPUs in data centers built for agent workloads.
The expected hardware mix illustrates the shift. The GPU-to-CPU ratio is projected to move from 8-to-1 for AI training and 4-to-1 for inference toward 1-to-1 in servers dedicated to AI agents.
The company announced the Arm AGI CPU in March, extending its business beyond licensing and platform technology into production processors. Arm later said the chip was being developed with ecosystem partners including Meta, OpenAI, Cloudflare, Oracle and SAP, while companies such as Google Cloud, Microsoft Azure and Nvidia were supporting agent workloads on Arm-based platforms.
AMD is already a major server-CPU supplier and has taken market share from Intel. The market estimate cited in the reporting puts AMD’s share at 27.7% this year and 30.7% in 2030, below the 50% share the company has said it is targeting. AMD also sells accelerators for AI inference, giving it exposure to both CPU demand and the computing used to execute trained models.
Supply constraints have supported revenue, including sales of older-generation processors and higher prices, but the same forecast projects Intel’s share falling to about 22% by 2030. Its participation in the opportunity is also tied to a loss-making foundry business that requires continued investment.
Nvidia’s CPU business is smaller than its graphics operation, but the company sells Arm-based CPUs on their own, alongside GPUs and within complete server racks. That structure gives Nvidia a route into the CPU demand associated with AI agents, while its GPUs and integrated data-center systems remain the larger parts of its business.
Analysis: The four companies offer different ways to participate in a possible agent-driven CPU expansion. Arm is positioned around processor designs and its own CPU platform, AMD combines server processors with AI accelerators, Intel starts from the largest existing CPU base but faces a projected share decline, and Nvidia adds CPUs to a much larger GPU and systems business. The investment outcome therefore depends on how broadly agent workloads are deployed and how much of that deployment uses each company’s products.
