Elon Musk has announced that Tesla will restart development of Dojo3, the company’s previously shelved third-generation AI chip project—this time with a dramatic shift in purpose. Rather than supporting Earth-based training for self-driving systems, Dojo3 will now be dedicated entirely to “space-based AI compute.” The announcement marks a sharp strategic reversal just five months after Tesla effectively shut down the Dojo initiative, dismantling its supercomputer team following the departure of Dojo lead Peter Bannon and the migration of roughly 20 engineers to DensityAI, a newly formed AI infrastructure startup.

At the time of the shutdown, Tesla appeared to be moving away from custom silicon, opting instead to rely more heavily on external partners such as Nvidia and AMD for compute, and Samsung for chip fabrication. Musk’s latest remarks, however, indicate a renewed confidence in Tesla’s internal chip roadmap and a willingness to pursue far more ambitious applications.

From Earth-Based Compute to an Orbital “Moonshot”

Musk revealed on X that the decision to resurrect Dojo was driven by progress on Tesla’s in-house chip development, noting that the AI5 chip design is “in good shape.” The AI5 chip, produced by TSMC, is intended to power Tesla’s Full Self-Driving software as well as its Optimus humanoid robots.

Looking further ahead, Tesla last summer secured a $16.5 billion deal with Samsung to manufacture its upcoming AI6 chips, which will support vehicles, robots, and large-scale AI training in data centers. Against this backdrop, Musk positioned the revived Dojo3—now referred to as AI7/Dojo3—as a long-term “moonshot” aimed at enabling space-based AI infrastructure.

To support this renewed push, Tesla is actively rebuilding the Dojo team that was recently disbanded. Musk has personally taken to X to recruit engineers, inviting candidates interested in designing “the highest volume chips in the world” to apply directly.

Space-Based AI and the Growing Off-Planet Compute Vision

The timing of Musk’s announcement is notable, arriving alongside Nvidia’s unveiling of Alpamayo, an open-source autonomous driving model introduced at CES 2026 that directly competes with Tesla’s Full Self-Driving system. While acknowledging the difficulty of solving rare and unpredictable driving edge cases, Musk publicly welcomed Nvidia’s progress, framing competition as part of a broader technological push.

Musk’s vision aligns with a growing belief among AI leaders that the future of large-scale compute may extend beyond Earth. With data centers placing increasing strain on terrestrial power grids, off-planet infrastructure has begun to attract serious interest. OpenAI CEO Sam Altman—despite being a rival—has reportedly expressed enthusiasm for orbiting data centers.

Musk’s control of SpaceX gives him a unique advantage in pursuing this concept. According to Axios, he intends to leverage SpaceX’s planned IPO to help fund a constellation of compute satellites deployed via Starship. Operating in constant sunlight, these satellites could theoretically harness uninterrupted solar power. Even so, significant technical hurdles remain, particularly the challenge of cooling high-density computing hardware in the vacuum of space.

As with many of Musk’s ventures, Tesla’s move toward space-based AI compute follows a familiar pattern: an audacious idea that borders on science fiction—paired with an attempt to will it into existence through sheer scale, capital, and engineering ambition.