At the OCP Global Summit 2026, Arm executives described how CPUs support AI data centers as agentic AI workloads expand. Arm projects the CPU market will grow from $52 billion in 2026 to $300 billion by 2030, citing research projections. The company debuted its Arm AGI CPU in March for agentic AI workloads.

Arm executive Mohamed Awad outlined three CPU roles: AI head nodes, cloud-native CPUs and agentic CPUs. Head nodes schedule work, move data and manage memory to keep costly accelerators productive. High-core-count custom cloud-native CPUs support efficient, scalable virtual machines, while custom DPUs handle networking, storage, security and control-plane functions. Awad described an AI agent as “a virtual user who never sleeps” and said dedicated CPUs, memory and other resources are needed for these workloads.

Ravi Malhotra said agentic AI infrastructure has two complementary parts: AI racks that run models for reasoning and planning, and CPU racks that run agentic tools and orchestration. AI infrastructure is judged by token throughput, including tokens per second, per watt and per dollar. Enterprise customers deploying agentic infrastructure assess how to pair AI infrastructure with CPU racks.

Arm said its AGI CPU is the company’s first server-grade CPU silicon. Malhotra said its design prioritizes performance, scalability and efficiency and uses the third-generation Neoverse core, developed with hyperscaler partners. Arm focused on reducing memory latency, providing high memory bandwidth per core and maintaining linear scalability, with performance sustained at 90% to 95% utilization. Malhotra said its power envelope can reduce cooling costs and allow more cores per rack.

At the OCP Global Summit 2026, Arm executives described how CPUs support AI data centers as agentic AI workloads expand. Arm projects the CPU market will grow from $52 billion in 2026 to $300 billion by 2030, citing research projections. The company debuted its Arm AGI CPU in March for agentic AI workloads.

Arm executive Mohamed Awad outlined three CPU roles: AI head nodes, cloud-native CPUs and agentic CPUs. Head nodes schedule work, move data and manage memory to keep costly accelerators productive. High-core-count custom cloud-native CPUs support efficient, scalable virtual machines, while custom DPUs handle networking, storage, security and control-plane functions. Awad described an AI agent as “a virtual user who never sleeps” and said dedicated CPUs, memory and other resources are needed for these workloads.

Ravi Malhotra said agentic AI infrastructure has two complementary parts: AI racks that run models for reasoning and planning, and CPU racks that run agentic tools and orchestration. AI infrastructure is judged by token throughput, including tokens per second, per watt and per dollar. Enterprise customers deploying agentic infrastructure assess how to pair AI infrastructure with CPU racks.

Arm said its AGI CPU is the company’s first server-grade CPU silicon. Malhotra said its design prioritizes performance, scalability and efficiency and uses the third-generation Neoverse core, developed with hyperscaler partners. Arm focused on reducing memory latency, providing high memory bandwidth per core and maintaining linear scalability, with performance sustained at 90% to 95% utilization. Malhotra said its power envelope can reduce cooling costs and allow more cores per rack.