Elon Musk Revives Tesla’s Dojo3 Chip Project, Targets Space AI

Jan 21, CFA News — Tesla CEO Elon Musk announced the revival of the previously shelved Dojo 3 supercomputer project, signaling a major shift in Tesla’s chip strategy. Unlike the August 2025 shutdown, Dojo3 will now extend beyond Earth-based autonomous driving model training to space-based artificial intelligence (AI) computing.

Dojo3 Revival

Five months ago, Dojo3 was abruptly halted, with the core team disbanded. Project lead Peter Bannon departed, and about 20 members founded DensityAI. Musk explained that developing two distinct AI chip designs was inefficient, and resources would focus on AI5 and AI6. Now, given AI5’s progress, Musk has restarted Dojo, issuing a recruitment notice for engineers interested in “producing the highest-volume chips in the world.”

Dojo3 Overview

Launched at Tesla AI Day 2021, the Dojo project was designed as a supercomputer for machine learning training. Dojo3 focuses on architecture restructuring and cost optimization, abandoning the complex D1 chip and wafer-level packaging of previous generations. AI5 chips, manufactured by TSMC, power Tesla vehicles and Optimus humanoid robots. AI6 chips, produced by Samsung under a $16.5 billion contract, support high-performance AI training in data centers. Experts note that Dojo3 could significantly accelerate the commercialization of FSD and Optimus if it optimizes AI training costs and efficiency.

Targeting Space AI

Musk stated that AI7/Dojo3 will be used for space AI computing, envisioning orbital deployment of large-scale AI systems within 4–5 years. He highlights cost advantages due to “free” solar energy and relatively simple cooling solutions, claiming that space-based AI could outperform terrestrial AI before Earth’s energy resources are depleted.

Challenges Ahead

Despite the ambitious plan, building space AI data centers faces multiple hurdles: orbital debris, regulatory approvals, and high-power cooling in vacuum. Temperature fluctuations vary across orbits: LEO (-65°C to +125°C), GEO (-20°C to +80°C). Megawatt-level GPU clusters require tens of thousands of square meters of radiative cooling panels. Launching hardware at this scale would need thousands of Starship flights, making a 4–5 year timeline extremely challenging and costly.