Elon Musk installed 100,000 Nvidia H200 accelerators in 19 days – it usually takes 4 years

The xAI team led by Elon Musk completed the installation of 100,000 Nvidia H200 Blackwell compute accelerators in just 19 days. According to the head of Nvidia, Musk completed what would normally take four years in record time, including building the building and setting up the equipment. All accelerators are part of the xAI supercomputer.

To understand the scale of the event, Nvidia CEO Jensen Huang explained that a typical data center takes about four years to complete such a task. In this case, three years are spent on planning, and the last year on delivery and installation of equipment. Huang expressed admiration for the speed and scale of the project, calling the efforts of Musk and his team “superhuman.” “Elon Musk is a superman. What would have taken others at least a year, he did in 19 days,” Huang noted. The process included, among other things, launching the first training run of xAI artificial intelligence on a newly built supercluster.

Huang also emphasized the complexity of his company’s network infrastructure, noting that connecting Nvidia equipment is much more complex than connecting traditional data centers: “The number of wires that go to one node and the back of the computer – it’s just all wires there…” According to him, the integration of 100,000 H200 video cards, implemented by Elon Musk, is an “unprecedented achievement” that is unlikely to be repeated by another company in the near future.