Elon Musk's SpaceXAI to add another 660,000 AI GPUs this year, nearing a total of 1.44 million in operation
Elon Musk says that the Colossus 2 will receive 220,000 GB300 GPUs by next week, with another two tranches of the same amount expected to arrive later this year. This will put the site at over a million AI GPUs, finally hitting the goal the billionaire set two years ago.
Elon Musk's SpaceXAI aims to add 660,000 AI GPUs to its Colossus supercomputer this year, nearly reaching a total of 1.44 million GPUs in operation. The billionaire announced the plan on X, stating that 220,000 Nvidia GB300 GPUs will be operational by next week, with another 220,000 coming online in November. Musk also hinted at an additional 220,000 units by late December, if they are lucky.
Colossus 1 currently features 150k H100, 50k H200, and 30k GB200 GPUs, while Colossus 2 boasts 110k GB200 and 440k GB300 GPUs. With these additions, SpaceXAI's GB300 GPU count will reach 1.1 million units, and the total GPU count in operation will be 1,440,000 units. This achievement is particularly notable as SpaceXAI is a comparatively younger company, having been established just three years ago, while industry rivals Anthropic and OpenAI have been operational for six and ten years, respectively.
Interestingly, the original Colossus 1 site, combining Hopper and Blackwell GPUs, proved inefficient for training Grok, leading Musk to rent it out to Anthropic for inference purposes. Conversely, Colossus 2 exclusively utilizes Blackwell GPUs, eliminating any potential bottlenecks. While securing these GPUs was a challenge, Musk's primary hurdle now lies in meeting the increasing power demands of his data centers.
To address this issue, SpaceXAI developed its own power infrastructure, resulting in some neighborhood disputes that culminated in a lawsuit for unauthorized gas turbines. The company is committed to rectifying this situation by decommissioning the turbines within a year, as its own 1.2-GW power plant is set to become operational.
While SpaceXAI's accomplishment is remarkable, it's not the only entity pursuing this goal. In 2024, Broadcom revealed that three hyperscale customers are targeting a million GPUs by 2027, though the identities of these customers remain undisclosed. Musk's ambition extends beyond the current target, with SpaceXAI planning to increase its data center capacity sevenfold by 2027 and aiming for 50 million H100-equivalent GPUs by 2030.
These data centers will not be confined to the ground, as SpaceX intends to launch an Orbital Data Center System featuring a million satellites, despite industry titan Jensen Huang deeming this concept a "dream" for now.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
