SpaceXAI plans to add 660,000 more GPUs, pushing its total toward 1.44 million by the end of 2026.
In a post on X, Elon Musk said 220,000 Nvidia GB300 GPUs will come online by early October 2026, another 220,000 in November 2026, and a further 220,000 by late December 2026 - that last batch only "if we get lucky." Those additions would stack on top of the 110,000 GB200 and 440,000 GB300 chips already running at the all-Blackwell Colossus 2 site, plus the 150,000 H100, 50,000 H200, and 30,000 GB200 GPUs at the older, mixed Colossus 1. If Musk hits every milestone, SpaceXAI's GB300 count alone reaches 1.1 million, and total GPUs across both sites hit 1.44 million.
That would put SpaceXAI within reach of a scale no rival has publicly claimed: over a million GPUs at a single company, built by a unit that is only three years old, versus six for Anthropic and ten for OpenAI. Getting the chips is apparently the easy part now - the harder problem is power, and SpaceXAI has been building its own 1.2-gigawatt plant to skip the queue everyone else is stuck in.
That power plant is also the subject of a lawsuit over unpermitted gas turbines, which the company has agreed to remove - over the next year, as its own generation comes fully online.