Musk Reveals SpaceXAI Compute Plans: Colossus 2 to Reach 1.1 Million Nvidia GB300 GPUs by End of 2026

On September 25, Elon Musk posted on X to disclose the compute deployment of his AI company, SpaceXAI. The company currently operates two major compute clusters, Colossus 1 and Colossus 2. Colossus 1 comprises 150,000 Nvidia H100, 50,000 H200, and 30,000 GB200 GPUs, while Colossus 2 has 110,000 GB200 and 440,000 GB300 GPUs.

Musk said SpaceXAI plans to bring 220,000 GB300 GPUs online in early October 2026, followed by another 220,000 in November 2026. If deployment goes smoothly, an additional 220,000 GB300 GPUs are expected by the end of December 2026, bringing Colossus 2 to 1.1 million GB300 GPUs by year-end. GB200 and GB300 are Nvidia's next-generation GPUs based on the Blackwell architecture, designed for AI training and inference, offering significant improvements in energy efficiency and compute density over the H-series.

Regarding competition, Musk noted that SpaceXAI has only three years of AI research history, while rivals such as Anthropic and OpenAI have accumulated six to ten years. He believes that if the current momentum continues, SpaceXAI could reach a leading position in about six months.