Colossus training cluster comes online with 100,000 NVIDIA H100 GPUs

xAI/Colossus

xAI's Colossus training cluster, built in Memphis over 122 days, went online with 100,000 H100 GPUs, making it the world's most powerful AI training system at launch. Elon Musk announced plans to double capacity to 200,000 GPUs within months, including 50,000 H200s, outpacing both Google's 90,000-GPU and OpenAI's 80,000-GPU clusters.

• Built from start to finish in 122 days at a former manufacturing site in Memphis, Tennessee.
• Expansion to 200,000 GPUs (including 50,000 H200s) planned within months of launch.
• Exceeds computing infrastructure of Google (90,000 H100s) and OpenAI (80,000 GPUs).
• Project cost estimated between $3–4 billion; hardware supplied by Dell and Supermicro.