Arm has expanded its data center lineup with Neoverse Compute Subsystems N4, its most configurable server blueprint yet. The launch targets the computing shift the company calls the agentic era, in which AI agents retrieve information, call software tools and coordinate accelerators instead of just generating text.
That workload mix pushes far more work onto the CPU and into the memory system, which is exactly where Arm says N4 is tuned. The subsystem supports up to 128 cores per die along with LPDDR6 memory and PCIe Gen 7 connectivity, delivering up to twice the performance, 1.25 times the performance per watt and 1.75 times the memory bandwidth of its predecessor N3.
Arm sells the subsystem as a pre-integrated foundation for custom silicon, letting partners build differentiated DPUs, networking parts and specialized compute platforms without validating every block themselves. The company is simultaneously pushing its ready-made Arm AGI CPU for customers that want to skip custom design entirely.
Momentum around that processor is building. OpenAI, Meta, Cloudflare, Oracle, SAP, Lenovo, Supermicro and Verda are developing solutions on the AGI CPU, and ByteDance’s Volcano Engine is preparing what Arm describes as the first agentic sandboxes powered by the chip. Google Cloud runs agent workloads on its Axion processors, while Microsoft Azure uses its Cobalt 200 parts for sandbox tool execution.
Arm argues that performance per watt, compute density and memory bandwidth will decide which server platforms win as agents multiply, and it wants one common Neoverse foundation underneath both custom and off-the-shelf designs.
