SpaceXAI is expanding its computing infrastructure by adding 660,000 AI GPUs throughout this year. This expansion will bring the company's total active GPU count to 1.44 million units. The fleet includes a mix of NVIDIA H100, H200, GB200, and GB300 models across two primary sites, Colossus 1 and Colossus 2.
The Colossus 1 facility currently houses 150,000 H100, 50,000 H200, and 30,000 GB200 units. Due to inefficiencies in training the Grok model, this site is being used by Anthropic for inference tasks. Colossus 2 is dedicated to Blackwell architecture, utilizing 110,000 GB200 and 440,000 GB300 units, with further batches of GB300s scheduled for deployment.
To support this massive hardware scale, SpaceXAI is constructing a 1.2-gigawatt power plant. The company previously faced a lawsuit regarding the use of unpermitted gas turbines at its facilities and has committed to decommissioning them within one year as the new power infrastructure becomes operational.
Looking ahead, the company plans to increase its data center capacity sevenfold by 2027. Elon Musk has stated an ambition to reach a total of 50 million H100-equivalent GPUs by 2030. Additionally, the firm is exploring the development of an Orbital Data Center System involving a constellation of one million satellites.
