Tesla completed the deployment of an AI cluster of 50,000 NVIDIA H100 AI accelerators in the fourth quarter of last year. In a presentation to shareholders, it was noted that the Cortex cluster was operational at Tesla’s Gigafactory facility in Austin, Texas, Datacenter Dynamics reports.
The information first appeared in the company’s Q4 and FY24 report. The new cluster is not related to the Dojo supercomputer, designed for FSD autonomous driving technologies, which has its own architecture and is equipped with custom D1 chips. However, the presentation dedicated to the report does not mention Dojo at all.
While the company has not specified when exactly in Q4 the rollout began, a Tesla spokesperson said on an October 2024 earnings call that the company was “on track to deploy 50,000 accelerators in Texas by the end of this month.” The project is reportedly behind schedule because Elon Musk fired the head of construction in April and ordered 12,000 H100 accelerators originally intended for Tesla to be transferred to xAI.
Image Source: Tesla
The presentation reports that Cortex has already helped create the FSD V13 (Supervised) “autopilot”. The new version has improved driving safety and comfort due to a 4.2-fold increase in data volume, an increase in video stream resolution, and other improvements. At the same time, the company announced the continuation of work on the software and hardware of the Optimus robot, including new-generation arms and movement mechanisms. Training was also carried out to perform additional tasks before the start of pilot production in 2025.
As for the company’s Q4 earnings, Elon Musk said in late January that the business continues to invest in training infrastructure outside of its Texas headquarters. Tesla was reportedly ramping up its computing power to train Optimus, which the billionaire said would require at least 10 times more resources to train Optimus than it would to fully train the car’s systems.
Tesla’s 2024 capital expenditures totaled $10 billion, and the company plans to spend the same amount over the next two years, though most of that will be on electric vehicle infrastructure. The company’s Q4 earnings report noted that its total AI-related capital expenditures, including infrastructure, exceeded $5 billion.