Tesla Dojo Resurrected, Aiming for a 'Space Data Center'—AI Supremacy Beyond Terrestrial Power
Tesla's proprietary AI computing infrastructure, 'Dojo,' has returned to the center of attention. Elon Musk has restarted the '3rd generation' Dojo (Dojo3), which had been effectively halted, and has indicated a policy of shifting its focus away from terrestrial autonomous driving training and toward space-based AI compute. This statement is also a 'pivot' that could once again overturn the flow of Tesla's chip strategy, talent drain, and external dependencies (NVIDIA/AMD/Samsung).
1. The 'Substance' of the Dojo3 Restart: From Terrestrial Autonomous Driving to 'Space Computing'
Musk posted on X that he would dedicate AI7/Dojo3 to space-based AI compute. Traditionally, Dojo was supposed to serve as a 'terrestrial training factory' for massive learning of in-vehicle camera footage to hone FSD (Full Self-Driving). However, this time it is being described as a 'moonshot' on the level of a lunar mission.
Even more symbolic is the 'direct recruitment' in the same post.
'If you want to work on the most high-volume chips in the world... (send me the 3 hardest technical problems you've solved)'—he recruited engineers on the premise of rebuilding the Dojo team.
2. It Was Disbanded 5 Months Ago: Why the Strategy Wavered
This is where the most drama lies. In the summer of 2025, it was reported that the Dojo team had been disbanded and the head, Peter Bannon, had resigned. The trend was strengthening for Tesla to rely on external sources like NVIDIA and AMD for computing resources and to utilize Samsung for manufacturing.
So, why the revival? Musk cites the progress of the internal chip roadmap, specifically that the AI5 design is in 'good shape,' as the basis. AI5 is manufactured by TSMC and is intended to support the vehicle's autonomous driving features and Optimus (humanoid robot).
Furthermore, the next chip, AI6, is with Samsung, with a manufacturing contract reported to be worth a total of $16.5 billion.
In short, rather than returning from 'buying outside' to 'making it,' it is more natural to interpret this as 'now that we have a prospect of being able to make it, we are returning Dojo (the training infrastructure) to in-house development.'
3. The Reality Posed by CES 2026: NVIDIA 'Open Sources' Autonomous Driving AI
The timing is also significant. At CES 2026, NVIDIA announced 'Alpamayo,' a suite of open-source AI models for autonomous driving development.
This can be read as a move to provide a 'common foundation' for other companies to catch up with 'Tesla's proprietary software (FSD).' In fact, Musk himself admitted that crushing the 'long tail of rare edge cases' is super hard, while also commenting that he 'wants them to succeed.'
In other words, the competitive environment is no longer a simple composition where 'only Tesla wins with massive data.' That is precisely why Tesla is presenting a narrative that extends its path to victory into 'space' while continuing the battle on the ground.
4. Is a Space Data Center a 'Dream Story'?: The Biggest Barrier is Cooling
The idea of a 'data center in space' is not unique to Musk. Against the backdrop of terrestrial power constraints and political risks, Axios reported on a blueprint where Musk (SpaceX) envisions an orbital data center, launching a 'computing satellite constellation' via Starship, and utilizing constant sunlight for 24/7 solar power generation. It further states that OpenAI's Sam Altman has also shown strong interest in this direction.
However, the biggest obstacle in reality is 'cooling.' In the vacuum of space, heat cannot be dissipated by air, and high-density computing is prone to failure due to heat. Axios also emphasizes this as the biggest 'reality check.'
Therefore, whether Dojo3 will truly operate in space remains unknown, but at the very least, this declaration is a message to investors and the recruitment market showing the 'range' of Tesla's AI, connecting it from 'cars' to 'robots,' 'data centers,' and finally 'space infrastructure.'