SpaceXAI plans 660,000 more AI GPUs in 2026 as Colossus nears 1.44 million units
SpaceXAI intends to bring 660,000 additional Nvidia GB300 GPUs online this year, in three batches of 220,000, according to Elon Musk. Those additions would push its total operating GPU count to roughly 1.44 million across Colossus 1 and Colossus 2. The company is also building a 1.2-gigawatt power plant to support the systems, following local disputes over unpermitted gas turbines.
Musk’s update breaks the 660,000-unit addition into three equal batches: 220,000 GB300s expected online within a week, another 220,000 in November, and a final 220,000 by late December if conditions allow. Existing capacity spans two sites: Colossus 1 runs 150,000 H100s, 50,000 H200s, and 30,000 GB200s, while Colossus 2 runs 110,000 GB200s and 440,000 GB300s.
Because Colossus 1 mixes Hopper and Blackwell generations, it is less suited for Grok training and has been leased to Anthropic for inference. Colossus 2 uses only Blackwell chips. SpaceXAI’s power constraints led to a dispute over unpermitted gas turbines; it has pledged to remove them over a year as a 1.2-GW plant is built.
The expansion could intensify pressure on local power infrastructure and air quality near the Colossus sites, affecting nearby residents and regulators already concerned about turbine permitting. It may also give SpaceXAI more compute for training and serving large models, potentially accelerating AI products and competition. The push for dedicated generation and the legal dispute over turbines may influence how future AI campuses are sited and approved, with implications for energy providers, technology firms, and host communities.