SpaceXAI to add 660,000 AI GPUs by year-end, reaching 1.44 million total

GPU Expansion Details

SpaceXAI is set to significantly expand its GPU capacity by deploying 660,000 new Nvidia GB300 GPUs by the end of 2024. This includes 220,000 units becoming operational next week, another 220,000 in November, and a final 220,000 by late December. This addition will bring the company's total GB300 GPUs to 1.1 million units and its overall GPU count to 1.44 million.

Current Infrastructure and Efficiency

The current GPU infrastructure at SpaceXAI includes 110,000 GB200 and 440,000 GB300 GPUs at Colossus 2, along with 150,000 H100, 50,000 H200, and 30,000 GB200 GPUs at Colossus 1. Colossus 1, which combines Hopper and Blackwell GPUs, has been deemed inefficient for training Grok and is rented to Anthropic for inference. Colossus 2, however, exclusively uses Blackwell GPUs to prevent bottlenecks in AI training.

Addressing Power Challenges

A major challenge for large-scale data center build-outs is power supply. SpaceXAI has addressed this by developing its own power solutions, including temporary use of unpermitted gas turbines, which led to a lawsuit. The company has committed to removing these turbines over a year as its own 1.2-GW power plant becomes operational.

Future Expansion Plans

Elon Musk has stated that the 1-million-GPU target is an initial step. SpaceXAI plans to increase its data center capacity sevenfold by 2027 and aims for 50 million H100-equivalent GPUs by 2030. Future plans also include an Orbital Data Center System with a million satellites, though this concept is currently considered a long-term aspiration.