Supercomputer Expansion
The end goal is finally in sight for Elon Musk, nearly two years after he announced plans to expand the Colossus supercomputer to over a million GPUs. The billionaire stated that 220,000 Nvidia GB300 GPUs will be operational soon, with another 220,000 coming online in November, and an additional 220,000 expected by late December.
Colossus 1 consists of 150,000 H100, 50,000 H200, and 30,000 GB200 units, while Colossus 2 houses 110,000 GB200 and 440,000 GB300 accelerators.
Hardware and Infrastructure
These deployments bring SpaceXAI’s total active hardware count to significant levels. The older Colossus 1 site, containing mixed Hopper and Blackwell architectures, has been utilized by Anthropic for inference tasks, whereas Colossus 2 relies entirely on Blackwell GPUs to avoid performance bottlenecks.
Power Challenges
Securing sufficient power remains a major hurdle for large-scale data center operators. SpaceXAI addressed this by developing its own power solutions, which previously led to regulatory disputes over unpermitted gas turbines. The company plans to phase these out as its permanent 1.2-gigawatt power plant comes online.
Future Targets
Industry competitors are also pursuing massive scale, with hardware partners noting multiple hyperscale customers targeting similar thresholds. SpaceXAI has outlined even larger ambitions for future growth, looking toward multi-million GPU configurations and exploring orbital data center concepts.
