It is thought that CEO Elon Musk poured about $10 billion (about 14 trillion won) into constructing computing infrastructure for artificial intelligence (AI) training this 12 months.
Twin Birch co-founder and Tesla investor Sawyer Merritt estimated through X (Twitter) on the thirty first (local time) that Tesla and xAI plan to operate $10 billion price of coaching computing capability by the top of this 12 months.
In keeping with this, in July, xAI began AI training using the Memphis supercluster ‘Colossus’, which integrates 100,000 ‘H100’ GPUs. This method requires an enormous amount of power, with 100,000 H100 GPUs alone consuming roughly 70 megawatts (MW), or no less than 150 MW overall.
The entire cost of the system is unknown, however the GPU alone is estimated to cost about $20,000 (roughly 28 million won), or roughly $2 billion (roughly 2 trillion won). GPUs typically account for half of the associated fee of constructing a knowledge center.
Also, Tesla unveiled the supercomputer ‘Cortex’ positioned in Texas at the top of August. This supercomputer is used to coach Tesla’s autonomous driving function ‘FSD’ and the humanoid robot ‘Optimus’.
Tesla announced that it plans to take a position $3 billion to $4 billion (about 4 trillion to five.5 trillion won) this 12 months alone to buy Nvidia AI chips. This place incorporates 100,000 NVIDIA ‘H100’ and ‘H200 chips’. It’s predicted that 130 megawatts (MW) of electricity will likely be needed this 12 months to power and funky the system, and that it will increase to greater than 500 MW over the following 18 months.
Along with Cortex, Tesla also has a supercomputer called ‘Dojo’ at its Buffalo, Latest York factory. A complete of $500 million (roughly 700 billion won) is planned to be moreover invested here.
Finally, xAI said it plans so as to add 50,000 H100 GPUs and 50,000 H200 GPUs to Colossus in the approaching months. The fee involved here amounts to billions of dollars.
Adding up the 4 cases above, the overall cost is well over $10 billion.
CEO Musk also announced through X last April, “Tesla is investing about $10 billion in AI for learning and inference this 12 months.” After all, this also appears to incorporate the associated fee of xAI.
After all, since it takes time for all AI servers to be installed and operate normally, the overall cost of AI hardware installed and operated by the 2 firms in 2024 is anticipated to be barely lower than $10 billion.
Reporter Park Chan cpark@aitimes.com
