Arm has declared a new turning point for artificial intelligence (AI) infrastructure by unveiling its first data center CPU that simultaneously ensures the cost-effectiveness and scalability of large-scale AI infrastructure, targeting the era of 'Agentic AI'.

On the 25th, Arm unveiled its first mass-produced silicon product, the data center CPU 'Arm AGI CPU', at the 'Arm Everywhere APAC' event.

Arm CEO René Haas emphasized in his keynote speech, “AI is fundamentally redefining how computing is built and deployed, and agentic computing is further accelerating this change.”

He characterized the launch of the AGI CPU as a significant turning point in Arm's history and explained the significance of platform expansion that extends from IP licensing to silicon.

Agentic AI refers to an environment that goes beyond simple model inference, where AI agents plan, execute, and operate continuously on their own.

In this process, the role of the CPU is rapidly expanding, including data movement, task coordination, and accelerator management.all.

To respond to these changes, Arm determined that a new data center CPU was needed that simultaneously meets high performance and high efficiency.

Arm AGI CPUs are designed based on Arm Neoverse V3 cores and feature up to 136 cores per CPU.

It provides up to 8,160 cores per rack in a high-density 1U server configuration, and can be expanded to over 45,000 cores in a liquid-cooled environment.

Arm explained that this enables it to provide more than double the performance per rack compared to existing x86-based systems.

CEO Haas stated, “AI data centers are being required to provide more computing power within the same power limits,” adding, “The Arm AGI CPU is designed to ensure both the economics and scalability of large-scale AI infrastructure based on a power-efficient architecture.”

This AGI CPU was created through joint development with Meta, and global companies such as OpenAI, SAP, SK Telecom, and Cloudflare are participating as initial partners.

Meta is pursuing gigawatt-class AI infrastructure optimization by combining its custom AI accelerator (MTIA) with AGI CPUs.

Arm also plans to release a reference server design based on Open Compute Project (OCP) standards and provide system architecture, firmware, and verification tools to the community.

The strategy is to lower adoption barriers across the entire ecosystem and accelerate the expansion of AI-native data centers.

CEO Haas stated, “Arm is no longer stopping at defining the architecture of AI-native data centers; it is building them directly,” adding, “AGI CPUs are part of Arm’s data center silicon roadmapHe stated that it was the starting point of.