Tesla has shared two images of some super sleek modern server cabinets with a Cyber design, taking styling from its Cybertruck. The clean wiring is next-level.
“Quiet please. The machines are learning,” said Tesla AI on X. Tesla did not specify what we are looking at, but it’s likely the company’s Dojo supercomputer setup for AI training of its Full Self-Driving (FSD) system, both designed in-house.
The cabinets are likely Dojo ExaPOD units, each containing trays of interconnected D1 chips (or possibly the next-gen D2 chips), providing massive compute power—potentially up to 1.1 exaflops per ExaPOD.
The clean, futuristic design, extensive cabling, and cooling infrastructure visible in the ceiling align with Dojo’s high power and thermal demands (e.g., 2.3 MW per cluster). We may be looking at Tesla’s Mojo Dojo Compute Hall in New York. It could also be the Dojo cluster being build at Giga Texas. The pictures match what Tesla shared last year of its Dojo 1 supercomputer.
Tesla CEO Elon Musk previously shared the company was operating 35,000 Nvidia H100 GPUs as of April 2024, with plans to increase this to 85,000 by the end of 2024, implying the purchase of an additional 50,000 H100s.
He estimated Tesla’s 2024 Nvidia spending at $3–4 billion, potentially equating to 75,000–133,000 GPUs, including various models, with 50,000 H100s allocated for Giga Texas to support Full Self-Driving (FSD) training.
Also, Musk noted that xAI (which recently purchased X) was using 100,000 H100 GPUs in its Memphis Supercluster as of July 2024 and planned to deploy a 300,000 B200 GPU supercomputer by summer 2025, indicating a further purchase of 200,000 B200 GPUs.
