Elon Musk is rushing to create the world's largest supercomputer

The construction process of Elon Musk's supercomputer facility is nearly complete, with machinery installation expected to begin in the next few months.

In May, The Information reported that the American billionaire and xAI company were planning to mobilize and connect 100,000 specialized graphics cards into a supercomputer called "Gigafactory of Computing", expected to operate in the summer. fall next year.

According to Tom's Hardware, Musk is working hard to get the system completed on schedule. He chose Supermicro to provide cooling solutions for both data centers containing Tesla and xAI supercomputers.

"Thank you, Musk, for taking the lead in choosing liquid cooling technology for giant AI data centers. This step can help preserve 20 billion trees for the planet," Charles Liang, founder and CEO of Supermicro, wrote on X on July 2.

Charles Liang shared a photo taken with Elon Musk at a data center. Photo: X/Charles Liang

AI data centers consume large amounts of electricity, Supermicro hopes their liquid solution will reduce electricity costs for infrastructure by up to 89% compared to traditional solutions using air.

In a previous tweet, Musk estimated that the Gigafactory supercomputer would consume 130 MW (130 million watts) of electricity when deployed. Incorporating Tesla's AI hardware after installation, the factory's power consumption can increase to 500 MW. The billionaire said the construction of the facility is almost complete, and machinery installation is expected to begin in the next few months.

Musk is building two parallel supercomputer clusters for Tesla and startup xAI, both estimated to be worth billions of dollars. According to Reuters, if 100,000 H100 GPUs are connected, the Gigafactory of Computing will be the most powerful supercomputer in the world, four times larger than the largest GPU cluster today.

xAI was founded by Elon Musk last July, competing directly with OpenAI and Microsoft. This startup then launched the Grok AI model, competing against ChatGPT. Earlier this year, Musk said training the next Grok 2 model would require about 20,000 Nvidia H100 GPUs, while Grok 3 could require 100,000 H100 chips.