Musk's SpaceXAI nears 1.44 million GPU milestone with three tranches of 220,000 GB300s
Elon Musk announced that SpaceXAI will deploy 660,000 additional Nvidia GB300 GPUs across three shipments this year, bringing the Colossus supercomputer complex to 1.44 million total processors in operation.
Nearly two years after unveiling his ambition to build out the Colossus supercomputer beyond one million GPUs, Elon Musk said the target is finally within reach. On X, the billionaire disclosed that 220,000 Nvidia GB300 GPUs will become operational within the next week, with an identical batch scheduled to come online in November and a third tranche of the same size arriving by late December, contingent on favorable circumstances.
Colossus 1 is 150k H100, 50k H200 and 30k GB200. Colossus 2 is 110k GB200 and 440k GB300. Another 220k GB300 will be fully operational next week and another 220k in November. If we get lucky, yet another 220k GB300 by late December.
Elon Musk, September 25, 2026
When combined with the 110,000 GB200 and 440,000 GB300 GPUs already running at Colossus 2, along with the 150,000 H100, 50,000 H200, and 30,000 GB200 units at Colossus 1, SpaceXAI's infrastructure will reach 1.1 million GB300 GPUs and 1,440,000 total processors. The achievement stands out given that SpaceXAI, founded only three years ago, is substantially newer than competitors like Anthropic and OpenAI, which have been operating for six and ten years respectively. Notably, Colossus 1, which combines Hopper and Blackwell architectures, proved unsuitable for Grok training, prompting Musk to lease the facility to Anthropic for inference workloads. By contrast, Colossus 2 relies exclusively on Blackwell GPUs, eliminating potential performance constraints.
Securing the GPUs themselves, despite Musk's earlier appeals to Nvidia chief Jensen Huang, has become less problematic than addressing infrastructure demands. Power availability represents the most pressing bottleneck for data center expansion, a challenge SpaceXAI addressed by constructing its own generation capacity. This approach sparked neighborhood opposition and resulted in litigation over unlicensed gas turbines. The company committed to decommissioning the turbines, though the process will extend over a year as a 1.2-GW power plant comes into service.
Operating a million or more GPUs at a single location represents a significant engineering milestone, yet SpaceXAI faces competition in pursuing this objective. Broadcom disclosed in 2024 that three hyperscale customers are targeting this capacity by 2027, though the company declined to identify them. For Musk, the one-million-GPU threshold marks merely an intermediate step. He has stated that SpaceXAI will expand its data center footprint sevenfold by 2027 and is pursuing an even more ambitious goal of 50 million H100-equivalent GPUs by 2030. The expansion extends beyond terrestrial facilities, with SpaceX planning an Orbital Data Center System featuring a million satellites, though Huang characterized the concept as a "dream" at present.