xAI & Tesla founder Elon Musk debuts Cortex AI supercluster, a first look at Tesla’s 50,000 Nvidia H100s to demonstrate the true demands of AI technology
Eager to continue delving into the world of supercomputing, Elon Musk is pressing forward with his newly renamed “Cortex” AI supercluster.
The so-dubbed technocrat has shared new information about the Tesla supercluster, which is arguably one of the world’s largest AI training clusters. It is said to contain 70,000 AI servers and will require 130 megawatts (MW) of cooling and power upon launch, which will then upscale to 500 MW by 2026.
Likewise, 50,000 Nvidia H100 enterprise graphics processing units (GPUs) are included within the cluster, alongside 20,000 of Tesla’s own hardware. Showcasing the above video on X, Elon Musk says the cluster is built to solve “real-world AI”.
What is the Cortex AI supercluster?
The supercluster is designed to enhance Tesla’s AI models for autonomous driving primarily, in addition to robotics, improved energy management and a broad range of other applications. Its development speaks to the growing demand of AI projects, with these powerful computing clusters seen as vital for advancing the advancement of AI technologies in growing industry areas.
First attracting attention in June 2024, large fans for Cortex were under construction to chill the entire supercluster. The fan stack is designed to cool the Supermicro-provided liquid cooling solution, which is built to handle an eventual 500 MW of cooling and power at full power.
This is a significantly large amount of power which will no doubt cause debate amongst global sustainability sectors, who are already concerned about the overall impact of such large-scale AI systems. Models like generative AI (Gen AI) in particular require significant levels of computational power and demand large amounts of data storage.
Key facts
- Cortex will contain 50,000 NVIDIA H100 GPUs and 20,000 units of Tesla’s custom AI hardware
- The system will require 130 MW of cooling and power (500 MW by 2026)
- The supercluster aims to enhance Tesla’s AI models for autonomous driving, energy management and more
As AI remains very power-hungry, plenty of work is required to ensure concerns around electricity consumption and a lack of cooling are mitigated so as to not overwhelm servers and avoid excess emissions.
Remaining a transformative force in the global business landscape, AI has real potential to drive important innovation. However, there is a fine balance to be trod when it comes to computing to avoid excess energy consumption.
The advancement of supercomputing
Cortex joins Elon Musk’s range of supercomputers currently in development. Already, he is involved in multiple large-scale AI computing projects, including the xAI Supercomputer that is planned to have 100,000 NVIDIA H100 GPUs, which will then upgrade to 300,000 NVIDIA B200 GPUs in 2025.
With Tesla, he is eager to train Tesla’s Self Driving (FSD) autopilot system, which will power consumer Teslas and the proposed ‘Cybertaxi’ product. This is in addition to training AI for Optimus, an autonomous humanoid robot that hasn’t yet entered production, but is expected to be used in Tesla’s manufacturing process.
From the advent of cloud computing in the early 2000s, there has been a significant shift in the capabilities of a data centre. Cloud data centres, operated by the likes of Amazon and Microsoft, have grown to become huge hubs of compute resource. They are able to offer scalable services to businesses worldwide, helping to lay the groundwork for the Gen AI revolution.
Now, supercomputing is a type of high-performance computing that uses multiple computers to work together and solve complex problems. Supercomputers are often used to solve problems that involve large amounts of data and billions or trillions of calculations.
