Elon Musk has announced Tesla's intention to restart development on Dojo3, the company's previously abandoned third-generation AI chip project. This time, however, the focus of Dojo3 will shift dramatically from training self-driving models on Earth to being dedicated entirely to "space-based AI compute." This significant strategic pivot comes just five months after Tesla had effectively halted its Dojo initiative, dismantling the supercomputer team following the departure of Dojo lead Peter Bannon and seeing approximately 20 Dojo workers move to DensityAI, a new AI infrastructure startup.

At the time of the initial shutdown, reports indicated Tesla planned to increase its reliance on external partners such as Nvidia and AMD for compute capabilities, and Samsung for chip manufacturing, rather than continuing with its custom silicon development. Musk's recent comments, however, suggest a clear re-evaluation of this strategy. He stated on X that the decision to revive Dojo was influenced by the positive status of Tesla’s in-house chip roadmap, specifically mentioning that the AI5 chip design was "in good shape." The AI5 chip, manufactured by TSMC, is designed to power Tesla's automated driving features and its Optimus humanoid robots.

Further elaborating on the company's chip strategy, last summer, Tesla secured a substantial $16.5 billion deal with Samsung for the production of its AI6 chips. These chips are intended to power Tesla vehicles, Optimus robots, and facilitate high-performance AI training within data centers. Musk explicitly positioned the resurrected Dojo3 project, now designated as "AI7/Dojo3," as a "moonshot" for space-based AI compute. To achieve this ambitious goal, Tesla is actively working to reconstitute the team that was disbanded just months prior. Musk personally used his X platform to recruit engineers, inviting those interested in developing "the highest volume chips in the world" to apply directly.

The timing of this announcement is particularly noteworthy, coinciding with Nvidia's unveiling of Alpamayo, an open-source AI model for autonomous driving at CES 2026, which directly competes with Tesla's Full Self-Driving (FSD) software. While acknowledging the immense difficulty of solving "the long tail of rare edge cases in driving," Musk expressed hope for Nvidia's success. This broader vision aligns with a growing sentiment among Musk and other AI executives, who argue that the future of data centers may increasingly be off-planet, citing the severe strain on Earth's existing power grids. OpenAI CEO Sam Altman, a rival of Musk, has also reportedly shown enthusiasm for the concept of orbiting data centers.

Musk possesses a unique advantage in pursuing this vision: his control over SpaceX's launch vehicles. According to Axios, he plans to leverage SpaceX's upcoming IPO to help fund his ambitious plan to deploy a constellation of compute satellites using Starship. These satellites would operate in constant sunlight, enabling continuous 24/7 solar power harvesting. Despite this advantage, numerous formidable challenges remain for establishing AI data centers in space, not least the complex problem of effectively cooling high-power compute equipment in the vacuum of space. Ultimately, Musk's announcement of Tesla embarking on "space-based AI compute" fits a familiar pattern: proposing a seemingly far-fetched idea and then attempting to forcefully bring it into reality.