SpaceX adds 660k AI GPUs nearing 1.44 million
Elon Musk’s Quest for a Million GPUs: The Race to AI Superintelligence
The ultimate objective for Elon Musk’s AI ambitions is finally within reach: scaling the Colossus supercomputer to over a million GPUs. This monumental undertaking signals a massive leap in computational power, positioning SpaceXAI at the forefront of the race to achieve true artificial intelligence.
The hardware delivery is already in motion. Musk has announced that 220,000 Nvidia GB300 GPUs will become operational next week, with another 220,000 units scheduled to come online in November. Even more exciting is the possibility of receiving another 220,000 units by late December, should the timeline prove favorable.
These massive additions dramatically boost the existing infrastructure. Combining the new deliveries with the current capacity—which already includes 110,000 GB200 and 440,000 GB300 GPUs in the Colossus 2 system, plus the 150,000 H100s and 50,000 H200s in Colossus 1—pushes SpaceXAI’s total operational GPUs to an astounding 1,440,000 units. This level of scale is a remarkable feat, especially considering that SpaceXAI is only three years old, while competitors like Anthropic and OpenAI have established themselves over a decade ago.
The architecture itself is designed for efficiency. The Colossus 2 site, for example, relies solely on Blackwell GPUs, ensuring maximum efficiency and avoiding computational bottlenecks. Meanwhile, the Colossus 1 site, featuring a mix of Hopper and Blackwell, was strategically rented out to Anthropic for inference work, demonstrating Musk’s ability to leverage unique compute resources.
This ambitious vision is fueled by a relentless pursuit of processing power. Beyond the immediate goal of a million GPUs, Musk is driving a much larger roadmap. He aims to grow data center capacity sevenfold by 2027, targeting 10 gigawatts of compute power and potentially reaching up to $500 billion in revenue by the end of next year. Furthermore, he envisions a future where AI compute is truly ubiquitous, even beyond Earth. There are plans to launch an Orbital Data Center System utilizing a million satellites to provide computational capability in orbit.
The long-term goal is set even higher: Musk is aiming for 50 million H100-equivalent GPUs by 2030, envisioning billions of GPUs as the foundation for future AI development. This quest for scale is not just about hardware; it involves solving immense engineering challenges, from powering massive facilities—which has led to public controversies over unpermitted gas turbines and environmental concerns—to figuring out how to harness space for computational power.
Buy on Amazon
Credit
