SpaceXAI plans 660,000 more Nvidia AI GPUs as Colossus 2 nears 1.44 million total
SpaceXAI intends to bring 660,000 additional Nvidia GB300 GPUs online this year, which would push its total operating GPU count to roughly 1.44 million. Colossus 2 already runs 110,000 GB200 and 440,000 GB300 chips, while Colossus 1 uses older H100, H200, and GB200 hardware. The company is also building a 1.2-gigawatt power plant and has faced a lawsuit over unpermitted gas turbines.
Colossus 1 combines 150,000 H100, 50,000 H200, and 30,000 GB200 chips. Colossus 2 currently has 110,000 GB200 and 440,000 GB300 units. Musk said 220,000 GB300s should be live next week, another 220,000 in November, and a final 220,000 by late December if conditions allow. That would bring total operational GPUs to about 1.44 million.
The company is only three years old, compared with six for Anthropic and ten for OpenAI. Because Colossus 1 mixes Hopper and Blackwell hardware and is inefficient for training Grok, it was rented to Anthropic for inference. Colossus 2 uses Blackwell exclusively. SpaceXAI is also building a 1.2-GW power plant after a lawsuit over unpermitted gas turbines.
The expansion could intensify demand for electricity and cooling infrastructure, affecting nearby residents through land use, emissions, and grid pressure. It may also accelerate AI model training and inference, influencing competition among AI firms and the availability of advanced computing. Workers and communities near data centers could see economic activity alongside environmental and noise concerns. Broader users may benefit from more capable AI services, though energy costs and local impacts may draw scrutiny.