Arm at the OCP Global Summit 2026: How Arm AGI CPU is scaling agentic AI infrastructure from CPU to rack

At the OCP Global Summit 2026 in San Jose, Arm highlighted how its Arm AGI CPU has moved from a product launch to real-world deployment, with production systems and partner traction validating its role in agentic AI infrastructure. The event emphasized open standards and collaboration, with Arm’s Open Agentic Infrastructure Platform offering modular, OCP-aligned designs to support scalable, flexible AI workloads. Mohamed Awad, Arm’s Executive Vice President and General Manager of Cloud AI, underscored a shift from human-initiated intent to machine-paced operations within data centers, driving new demands on CPU roles and system design.

A new methodology for measuring agentic workloads, developed with Fujitsu, Meta, and OpenAI, was introduced to evaluate CPU performance beyond traditional metrics, focusing on sustained efficiency in multistep tasks. Ravi Malhotra of Arm presented on rethinking CPU performance for agentic AI, noting that different agent types—such as software development agents or enterprise automation tools—impose distinct demands on infrastructure, including per-core performance and concurrency.

Independent analysis by Signal65 found Arm AGI CPU delivered an average of 1.99x modeled performance per rack compared to AMD EPYC 9755 across seven representative agentic AI workloads, based on verified server-level measurements. The findings highlighted the importance of compute density and efficiency when scaling agentic AI infrastructure, reinforcing the need for heterogeneous computing approaches.

Demonstrations at Arm’s booth illustrated agentic workflows in customer service applications, showing how AI agents execute multistep tasks rather than just generating responses. The event emphasized the growing role of heterogeneous computing, where CPUs, accelerators, memory, and interconnects must work together to optimize performance for agentic AI workloads.