Supermicro and Arm's AGI CPU delivers 2x rack performance and $10B savings
New Arm AGI CPU with 136 cores enables 6,000 cores per rack, cutting energy use by half.
Supermicro and Arm have unveiled a new class of energy-efficient rack-scale infrastructure specifically built for enterprise agentic AI. The Arm AGI CPU features a dense 136-core microarchitecture optimized for performance, delivering 6 GB/s memory bandwidth per core and latency-optimized memory access. These servers enable over 2x performance per rack compared to traditional architectures, with up to 6,000 cores in a single air-cooled rack. The lineup includes five models: a 2U Hyper server for memory-intensive workloads, a 5U GPU server with support for up to eight double-width GPUs, and a liquid-cooled multi-node solution for high-density deployments.
Supermicro's DCBBS technology stack and global manufacturing reduce time-to-online for large-scale AI deployments. Arm estimates that enterprises can save up to $10 billion in CAPEX per Gigawatt of AI data center capacity. The combination of Arm's energy-efficient Neoverse CSS V3-based CPUs and Supermicro's modular designs maximizes performance-per-watt and lowers energy demand, helping enterprises scale agentic AI without massive infrastructure costs. This collaboration addresses the growing need for efficient orchestration of thousands of parallel AI tasks in modern data centers.
- Arm AGI CPU features 136 cores with 6GB/s memory bandwidth per core, enabling 6,000 cores per air-cooled rack.
- Supermicro's new servers deliver over 2x performance per rack compared to traditional architectures.
- Five server models available: 2U Hyper, 5U GPU (up to 8 GPUs), and liquid-cooled multi-node for high-density deployments.
Why It Matters
Enterprises can deploy agentic AI at scale with lower energy costs and up to $10B CAPEX savings per Gigawatt.