We love getting a peek inside a supercomputer. One of our recent favorites is Nvidia’s look at Eos, the ninth fastest supercomputer in the world.
Elon Musk has now given a sneak peek at the massive AI supercluster recently dubbed Cortex, which is being used by X (formerly Twitter).
The supercluster, currently under construction at Tesla’s Giga factory in Texas, is expected to house 70,000 AI servers. Its initial power and cooling requirements are 130 megawatts, with a target of scaling to 500 megawatts by 2026.
Tesla’s AI Strategy
In the video, embedded below, Musk shows off rows and rows of server racks, which could potentially hold up to 2,000 GPU servers — a mere fraction of the 50,000 Nvidia H100 GPUs and 20,000 Tesla hardware units eventually expected to populate Cortex. The video, while brief, offers a rare inside look at the infrastructure that will soon power Tesla’s most ambitious AI projects.
Video from inside Cortex today, the massive new AI training supercluster being built at Tesla HQ in Austin to solve real-world AI problems pic.twitter.com/DwJVUWUrb5August 26, 2024
Cortex is being developed to enhance Tesla’s AI capabilities, particularly for training the Full Self-Driving (FSD) autopilot system used in its cars and the Optimus robot, an autonomous humanoid scheduled for limited production in 2025. The supercluster’s cooling system, featuring massive fans and liquid cooling provided by Supermicro, is designed to handle the extensive power demands that, Tom’s Hardware is comparable to a large coal-fired power station.
Cortex is part of Musk’s broader strategy to deploy multiple supercomputers, including the operational Memphis Supercluster, which runs on 100,000 Nvidia H100 GPUs, and the upcoming $500 million Dojo supercomputer in Buffalo, New York.
Despite some delays in upgrading to Nvidia’s latest Blackwell GPUs, Musk’s aggressive acquisition of AI hardware shows how eager Tesla is to be at the forefront of AI development.
The divided billionaire said earlier this year that the company planned to spend “over a billion dollars” on Nvidia and AMD hardware this year alone, just to stay competitive in the AI sector.
