Cerebras Unveils Nexus Rack-Scale Architecture for AI Hardware
Cerebras introduces Nexus, a new rack-scale architecture designed to decouple power, compute, and I/O to enable annual increases in AI speed. The modular design allows independent upgrades for power supplies, wafer-scale engine compute backpacks, and edge-connected I/O silicon without requiring full system redesigns. The CS-4, CS-5, and CS-6 systems will utilize the Nexus architecture to support multi-generation inference scaling. Open article →