Podcast Episode
Tesla Revives Abandoned Dojo3 Chip for Space-Based AI Computing
January 21, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
Elon Musk announced over the Martin Luther King Jr. holiday weekend that Tesla will resurrect its previously abandoned Dojo3 artificial intelligence chip project, pivoting the technology from training self-driving systems on Earth to powering orbital AI data centers. The announcement represents a dramatic strategic reversal just five months after Tesla disbanded its Dojo supercomputer team and reassigned engineers to other projects.
Musk posted on X on January 18, 2026, that AI7 slash Dojo3 will be for space-based AI compute, describing the resurrected project as a moonshot endeavour. The decision to revive Dojo was based on progress in Tesla's broader chip roadmap, with Musk noting that the AI5 chip design was in good shape and allowing the company to refocus resources on the more ambitious Dojo3 project.
At the time of the shutdown, Tesla signalled plans to deepen its reliance on NVIDIA and AMD for computing resources and Samsung for chip manufacturing. The company's 16.5 billion dollar deal with Samsung for AI6 chips, which will support vehicles, robots, and data centre training, remains intact. Tesla's AI5 chip, manufactured by TSMC, currently powers the company's Full Self-Driving features and Optimus humanoid robots.
An orbital data centre operating in a sun-synchronous orbit could theoretically harvest solar power continuously, eliminating downtime and reducing dependence on traditional energy grids. The vacuum of space also provides natural cooling advantages compared to terrestrial facilities, where cooling represents one of the largest operational costs.
Musk holds a distinct advantage through SpaceX, which controls the launch vehicles necessary for such a project. SpaceX is planning a 2026 initial public offering that could raise 25 billion dollars or more, with market estimates putting the company's post-IPO valuation at around 1.5 trillion dollars. Musk plans to leverage this IPO to fund a constellation of compute satellites launched by Starship rockets.
The economics remain speculative at this stage. Rocket launch costs alone pose a significant challenge, not to mention the need to replace onboard chips every 5 to 6 years. Building orbital data centres competitive with Earth-based facilities would require a very large number of launches. Analysts at Via Satellite characterized the near-term economics as speculative, noting that orbital data centres may initially suit only narrow use cases such as batch training and defence applications.
Musk believes running large-scale AI systems in orbit could become more cost-effective than terrestrial alternatives within 4 to 5 years, though this timeline remains to be validated.
The convergence of aerospace and technology industries around orbital computing gained fresh momentum at the AIAA SciTech Forum in Orlando, where Rob DeMillo, CEO of Sophia Space, presented the company's modular tile design for orbital computing using passive cooling systems.
The timing is notable given NVIDIA's unveiling of Alpamayo at CES 2026, an open-source autonomous driving AI model that directly challenges Tesla's proprietary Full Self-Driving software. Musk acknowledged on X that solving rare driving scenarios is super hard, adding that he honestly hopes they succeed.
Whether Musk's vision of orbital AI computing transitions from posts on social media to operational satellites remains to be seen, but the revival of Dojo3 signals that Tesla is committed to exploring what could become a transformative infrastructure shift for the artificial intelligence industry.
The Original Dojo Project and Its Abandonment
Tesla shut down the original Dojo effort in August 2025 following the departure of project leader Peter Bannon, who had joined the company from Apple in 2016. Approximately 20 team members subsequently joined DensityAI, a startup founded by former Dojo head Ganesh Venkataramanan alongside ex-Tesla employees Bill Chang and Ben Floering.At the time of the shutdown, Tesla signalled plans to deepen its reliance on NVIDIA and AMD for computing resources and Samsung for chip manufacturing. The company's 16.5 billion dollar deal with Samsung for AI6 chips, which will support vehicles, robots, and data centre training, remains intact. Tesla's AI5 chip, manufactured by TSMC, currently powers the company's Full Self-Driving features and Optimus humanoid robots.
The Case for Orbital Data Centres
The space-based AI compute concept addresses mounting energy constraints facing terrestrial data centres. Data centres and AI consumed 460 terawatt hours of electricity in 2022, and industry projections estimate this will more than double to over 1,000 terawatt hours by 2026. Musk and other AI executives, including OpenAI CEO Sam Altman, have expressed concern that Earth's power grids are already strained to capacity by the energy demands of modern AI training.An orbital data centre operating in a sun-synchronous orbit could theoretically harvest solar power continuously, eliminating downtime and reducing dependence on traditional energy grids. The vacuum of space also provides natural cooling advantages compared to terrestrial facilities, where cooling represents one of the largest operational costs.
Musk holds a distinct advantage through SpaceX, which controls the launch vehicles necessary for such a project. SpaceX is planning a 2026 initial public offering that could raise 25 billion dollars or more, with market estimates putting the company's post-IPO valuation at around 1.5 trillion dollars. Musk plans to leverage this IPO to fund a constellation of compute satellites launched by Starship rockets.
Technical and Economic Challenges
However, experts note formidable technical obstacles to realising orbital data centres. Heat dissipation in the vacuum of space requires entirely different thermal management systems compared to air-cooled terrestrial facilities. Components must be radiation-hardened to withstand cosmic rays and solar radiation that would quickly degrade standard electronics. Achieving reliable high-bandwidth communication with Earth adds another layer of complexity.The economics remain speculative at this stage. Rocket launch costs alone pose a significant challenge, not to mention the need to replace onboard chips every 5 to 6 years. Building orbital data centres competitive with Earth-based facilities would require a very large number of launches. Analysts at Via Satellite characterized the near-term economics as speculative, noting that orbital data centres may initially suit only narrow use cases such as batch training and defence applications.
Musk believes running large-scale AI systems in orbit could become more cost-effective than terrestrial alternatives within 4 to 5 years, though this timeline remains to be validated.
An Emerging Industry Trend
Tesla is not alone in pursuing space-based computing infrastructure. Google announced Project Suncatcher with plans to launch two test satellites carrying AI processing chips in 2027 in partnership with Planet Labs. Blue Origin has assembled a dedicated team working on creating technology to support AI data centres in space. Starcloud, an NVIDIA-backed startup, launched a satellite carrying an NVIDIA H100 graphics processing unit in December 2025.The convergence of aerospace and technology industries around orbital computing gained fresh momentum at the AIAA SciTech Forum in Orlando, where Rob DeMillo, CEO of Sophia Space, presented the company's modular tile design for orbital computing using passive cooling systems.
Rebuilding the Team
Tesla is now actively recruiting engineers to rebuild the capability it dismantled months ago. Musk directed interested candidates to email AI underscore Chips at Tesla dot com with 3 bullet points on the toughest technical problems you've solved.The timing is notable given NVIDIA's unveiling of Alpamayo at CES 2026, an open-source autonomous driving AI model that directly challenges Tesla's proprietary Full Self-Driving software. Musk acknowledged on X that solving rare driving scenarios is super hard, adding that he honestly hopes they succeed.
Whether Musk's vision of orbital AI computing transitions from posts on social media to operational satellites remains to be seen, but the revival of Dojo3 signals that Tesla is committed to exploring what could become a transformative infrastructure shift for the artificial intelligence industry.
Published January 21, 2026 at 2:11pm