ELON MUSK: 1M NVIDIA GPUS? NAH. MY SUPERCOMPUTERS NEED THE POWER OF 50M

Elon Musk has once again raised eyebrows in the tech world, claiming that the next generation of supercomputers needed to power his artificial intelligence ambitions would require the equivalent of 50 million Nvidia GPUs—a figure that dwarfs current global deployment levels and illustrates the scale at which Musk envisions AI development evolving.

- 

Speaking at a closed investor meeting for xAI, his AI startup launched in 2023, Musk reportedly downplayed recent efforts by other firms that boast access to 1 million Nvidia chips, saying, “That’s nothing. To do this right, you need the power of 50 million.” The remark, first reported by The Information, highlights Musk’s relentless push for computing supremacy amid an intensifying global race to develop more powerful AI systems.

BEYOND SCALE: MUSK’S AI INFRASTRUCTURE VISION

Musk’s comments reflect a growing belief that current compute capacity is vastly insufficient for the level of artificial general intelligence (AGI) he envisions. His startup xAI is already building a massive AI training cluster powered by Nvidia's H100 and next-gen B200 GPUs, but according to Musk, this is just the beginning.

He has previously stated that xAI would need “at least 100,000 H100s” to train its next-generation models—placing it among the most aggressive chip procurement plans in the industry. But even that, Musk suggests, is merely a stepping stone toward building an AI supercomputer infrastructure that rivals the scale of national power grids.

His vision calls for sprawling data centers packed with cutting-edge silicon, backed by advanced cooling systems and powered by low-carbon energy sources such as solar, nuclear, or hydrogen. In Musk's view, the compute needed to reach advanced, multi-modal AGI capable of reasoning, code synthesis, and autonomous planning at human or superhuman levels will exceed anything currently in production or planning.

A SHOT AT THE COMPETITION