Overview

• xAI's Colossus supercomputer, located in Memphis, Tennessee, currently operates with 100,000 NVIDIA Hopper GPUs and is recognized as the largest AI training cluster globally.
• The supercomputer was constructed in just 122 days, with AI model training commencing 19 days post-installation, a timeline praised as 'superhuman' by NVIDIA CEO Jensen Huang.
• Colossus utilizes Nvidia's Spectrum-X Ethernet for high data throughput and minimal latency, crucial for AI training tasks that require processing vast datasets efficiently.
• Plans are underway to expand Colossus to 200,000 GPUs, incorporating next-gen H200 GPUs alongside existing H100 units, which will significantly increase its computational capacity.
• The supercomputer supports xAI's Grok language models, aimed at enhancing AI-driven chatbot features for X Premium subscribers, highlighting the system's role in advancing AI technology.