xAI's Colossus project plans substantial GPU capacity expansion to support advanced model development, signaling accelerated AI compute infrastructure buildout by SpaceX/Elon Musk's entities.
Elon Musk, CEO of Space Exploration Technologies Corp., expressed confidence in SpaceX's AI capabilities, predicting the company will achieve a leadership position in the sector within six months. In a post on X on September 24, Musk stated he was "cautiously optimistic that SpaceX will have a Fable/GPT-6 level model in 2 to 3 months."
Elaborating on the company's trajectory, Musk emphasized rapid acceleration, noting that SpaceX's AI efforts span only three years compared to six and ten years for Anthropic and OpenAI respectively. "If our second derivative remains strong, SpaceX will reach pole position in about 6 months," he wrote.
Musk also addressed the relationship between computational capability and practical application, arguing that beyond a certain intelligence threshold, additional capability offers diminishing returns for specific tasks. He illustrated this point by suggesting that a "Newton-level" intellect in a toaster would be superfluous. He stressed that rapidly deploying massive computing infrastructure presents severe logistical challenges, citing SpaceX's competency in bringing large-scale hardware online as a competitive advantage.
The expansion of xAI, now known as SpaceXAI, underscores these ambitions. The company's Colossus superclusters, powered by Nvidia Corp.'s GPUs, now total approximately 670,000 units across two configurations. Colossus 1 comprises 150,000 H100 GPUs, 50,000 H200 units, and 30,000 GB200 units. Colossus 2 contains 110,000 GB200 and 440,000 GB300 units. Expansion plans are aggressive: 220,000 additional GB300 units were scheduled to become fully operational by late September, with another 220,000 coming in November and potentially a third batch of 220,000 by late December. Over its three-year history, SpaceXAI has released successive versions of its Grok model, including variants for image generation (Imagine), voice interaction, code generation (Build), and bot functionality, with version 4.7 representing the latest iteration.