Cerebras Systems — Condor Galaxy AI Supercomputer

The Condor Galaxy AI supercomputer is a joint venture between Cerebras Systems and Abu Dhabi-based G42, designed to provide an alternative to GPU-based AI training. Condor Galaxy uses Cerebras’s Wafer Scale Engine (WSE-2) chips — the world’s largest computer chips, each the size of an entire silicon wafer — which can train AI models faster than equivalent GPU clusters for certain workloads due to dramatically higher memory bandwidth. Condor Galaxy 1 (CG-1), deployed at a Colovore data center in Dallas, delivers 4 exaFLOPS of AI compute using 54 CS-2 systems. Additional sites are planned under a $900 million commitment from G42. Cerebras offers its compute via API for model training and fine-tuning, competing with AWS, Azure, and Google TPU cloud services.