AI

Arm Expands Compute Platform to Production Silicon with First Data Center CPU for Agentic AI

Arm announced that it is extending its compute platform to include production silicon products for the first time in the company's history. The expansion adds Arm-designed silicon alongside existing IP and Arm Compute Subsystems (CSS), providing partners with options across all three categories. The announcement introduces the Arm AGI CPU, an Arm-designed data center CPU targeted at agentic AI infrastructure. It is the first such silicon product from Arm and is intended to address workloads involving AI agents that reason, plan, and act, which increase token generation and require additional CPU capacity for reasoning, coordination, and data movement.
 
The rise of agentic AI is expected to drive data centers to require more than 4x the current CPU capacity per GW while operating within existing power constraints. The Arm AGI CPU is positioned as a response, offering a simplified architecture without x86 overhead and complexity.
 
Key specifications of the Arm AGI CPU include up to 136 Arm Neoverse V3 cores per CPU, 6GB/s memory bandwidth per core at sub-100ns latency, and a 300-watt TDP. It supports deterministic performance under sustained load through a dedicated core per program thread, avoiding throttl...
You've read this far — sign in to keep reading

Sign in to keep reading.

Forgot password?
OR