Supermicro reveals AI-centric solutions with Arm AGI CPUs

Super Micro Computer has announced a class of AI-focused products featuring Arm AGI CPUs. The increase in computational requirements for modern AI has led to a need for rack-scale infrastructure designed to optimise compute performance while operating within power and space constraints.

Supermicro’s solutions are intended to support the growth of AI workloads by improving performance, efficiency, and system density in rack-scale deployments. Its Data Center Building Block Solutions (DCBBS) are designed to reduce time-to-online for new infrastructure deployments.

The newly introduced platforms include air-cooled dual-socket 2U compute-optimised servers, 5U GPU-optimised rackmount servers, and a liquid-cooled multi-node system designed for agentic AI workloads. These systems integrate Arm Neoverse CSS V3-based CPUs and are designed to provide scalable infrastructure with improved performance-per-watt and reduced energy consumption.

When configured with Arm AGI CPUs, Supermicro systems are described as delivering more than twice the rack performance compared to traditional setups. According to Arm’s figures, this approach could reduce capital expenditure by up to $10 billion per gigawatt of AI data center capacity.

The Arm AGI CPU is described as having a 136-core architecture intended to improve performance and reduce legacy overhead, increasing task throughput per cycle. It includes 6 GB/s memory bandwidth per core and latency-optimised memory access to support scaling, along with expanded memory capacity and flexible I/O for distributed AI workloads.

The Supermicro Arm-based server lineup includes five models:

• 2U Hyper Server: designed for cloud and memory-intensive workloads; supports two Arm AGI CPUs and up to two GPUs.
• 5U GPU Server: designed for AI training workloads; supports up to eight GPUs.
• 2U4N Liquid-Cooled Server: designed for OCP ORV3 environments with multi-node configurations.
• 2U Hyper-E Server: single-socket, edge-focused design with flexible I/O options.
• 1U 4N in ORW Rack: high-density compute configuration for AI workloads.

Overall, the portfolio is positioned for use in AI data centre environments requiring scalable rack-scale infrastructure.