Elon Musk and his xAI team have pulled off what many thought impossible, setting up a colossal supercluster of 100,000 NVIDIA H200 GPUs in just 19 days. This feat is jaw-dropping, considering that building such an infrastructure typically takes several years. Jensen Huang, the CEO of NVIDIA, praised Musk’s team, describing the achievement as “superhuman” and emphasizing the complexity involved in building and networking NVIDIA’s hardware.

The scale of this project is massive. Not only did they need to set up thousands of GPUs, but they also had to construct an entire facility equipped with liquid cooling and immense power capacity to support the processing demands of xAI’s AI models. Huang shared that in a normal scenario, building a supercomputer of this magnitude would involve three years of planning and another year for installation and setup.

To put this into perspective, Huang explained how complex networking is with NVIDIA’s systems, noting that traditional data center hardware pales in comparison to the intricacies of managing this scale of GPUs. This supercluster, part of xAI’s “Colossus” system, is now one of the fastest supercomputers on the planet, powering AI research and projects that are expected to push boundaries in machine learning.

What makes this more impressive is that it was all coordinated between Musk’s engineering teams and NVIDIA’s experts, showcasing not only the ambition but also the technical prowess of both sides.