Five months ago, Elon Musk called Dojo an “evolutionary dead end” and disbanded the entire team. Now he wants to put it in orbit.

The whiplash is real. Over the long weekend, Musk announced on X that Tesla aims to restart work on Dojo3, its previously abandoned third-generation AI chip. The twist? This time it won’t train self-driving models on Earth. Instead, Musk says it will be dedicated to “space-based AI compute.”

The quick breakdown:

The Fact: Tesla is reviving its Dojo supercomputer project for orbital AI data centers, not autonomous driving.
The Delta: This announcement came just two weeks after Nvidia unveiled Alpamayo at CES 2026, an open-source autonomous driving AI that directly challenges Tesla’s Full Self-Driving.
The Buyer Impact: Tesla investors should understand that Dojo3 has nothing to do with making your car drive better. It is a SpaceX play disguised as a Tesla announcement.

The Dojo Shutdown Was Brutal and Recent

Tesla disbanded its Dojo supercomputer team in August 2025, marking a major retreat from years of ambitious in-house chip development. The company’s Dojo lead, Peter Bannon, departed following Musk’s order to shut down the program entirely. Around 20 team members left to join DensityAI, a startup founded by former Dojo head Ganesh Venkataramanan and ex-Tesla employees Bill Chang en Ben Floering.

At the time of shutdown, Bloomberg reported that Tesla planned to increase its reliance on Nvidia and partners like AMD for compute and Samsung for chip manufacturing rather than continue developing its own custom silicon. Musk posted on X that he was shutting down Dojo entirely, calling the project “an evolutionary dead end.”

The pivot left New York officials scrambling. As we documented in our coverage of Tesla’s $1 billion Buffalo boondoggle, the state had negotiated a lease extension predicated on Tesla’s $500 million Dojo investment. Officials learned about the shutdown through Musk’s social media announcement, not direct communication.

Dojo3 Resurrection Hinges on Tesla’s AI Chip Roadmap

Tesla’s decision to revive Dojo was based on the state of its in-house chip roadmap, with Musk noting that the company’s AI5 chip design was “in good shape.” The AI5 chip, manufactured by TSMC, was designed to power Tesla’s automated driving features and Optimus humanoid robots. Production will initially occur in TSMC’s Taiwan factory before transitioning to its Arizona facility.

Last summer, Tesla signed a $16.5 billion deal with Samsung to build its AI6 chips at the company’s new Taylor, Texas facility. The contract runs through 2033 and positions Samsung as a key production partner for Tesla’s custom-designed AI hardware. These chips are central to Tesla’s autonomous driving software, neural network training, and next-gen robotics.

“AI7/Dojo3 will be for space-based AI compute,” Musk said on Sunday, positioning the resurrected project as more of a moonshot than an evolution of its original purpose.

Tesla Now Recruiting Engineers It Fired Months Ago

To achieve its space-based AI ambitions, Tesla is gearing up to rebuild the team it dismantled months ago. Musk used the same post to recruit engineers directly, writing:

“If you’re interested in working on what will be the highest volume chips in the world, send a note to AI_Chips@Tesla.com with 3 bullet points on the toughest technical problems you’ve solved.”

The irony is thick. Many of the engineers with the most relevant experience already left for DensityAI or scattered across the industry following the August shutdown. Tesla’s talent exodus throughout 2025 included not just the Dojo team but also Milan Kovac (head of Optimus robotics engineering), David Lau (VP of software engineering), and Omead Afshar (Musk’s former chief of staff).

Nvidia’s Alpamayo Explains the Timing

The announcement’s timing is notable. At CES 2026, Nvidia unveiled Alpamayo, an open-source AI model family for autonomous driving that directly challenges Tesla’s Volledig zelfrijdend software. CEO Jensen Huang called it “the ChatGPT moment for physical AI,” positioning Alpamayo as the foundation for safe, scalable autonomy that any automaker can build upon.

Musk commented on X that solving the long tail of rare edge cases in driving is “super hard,” adding: “I honestly hope they succeed.” The measured response contrasts with his typical combative stance toward competitors. It suggests Tesla may be acknowledging that its autonomous driving edge is narrowing while pivoting its narrative toward space infrastructure.

Mercedes-Benz is already shipping the 2025 CLA with Nvidia’s full autonomous vehicle stack, including Alpamayo reasoning capabilities. If legacy automakers can suddenly field Tesla-competitive driver assistance using off-the-shelf Nvidia tools, Tesla’s differentiation in autonomy becomes a commodity rather than a moat.

Space Data Centers: The SpaceX Connection

Musk and several other AI executives have argued the future of data centers may lie off-planet, since Earth’s power grids are already strained to the max. Axios recently reported that Musk rival and OpenAI CEO Sam Altman is also excited by the prospect of putting data centers into orbit.

Musk has an edge over his peers because he already controls the launch vehicles through SpaceX. Per Axios, Musk plans to use SpaceX’s upcoming IPO to help finance his vision of using Starship to launch a constellation of compute satellites that can operate in constant sunlight, harvesting solar power 24/7.

SpaceX’s IPO, reportedly targeted for mid-to-late 2026, could raise upwards of $30 billion and value the company at $1.5 trillion. The proceeds would fund data centers in space built on modified Starlink satellite platforms.

This context reframes the Dojo3 announcement entirely. Tesla shareholders are being told this is about Tesla’s AI future. The reality? It is a public relations bridge connecting Tesla’s chip development to SpaceX’s orbital infrastructure ambitions.

The Technical Roadblocks Are Enormous

Still, there are many roadblocks to making AI data centers in space a possibility, not least the challenge of cooling high-power compute in a vacuum. Thermal management that works on Earth fails spectacularly in space, where there is no air to carry heat away. Every watt of compute generates heat that must be radiated into the void.

Musk’s comments about building “space-based AI compute” fit a familiar pattern: float an idea that sounds far-fetched, then try to brute-force it into reality. Sometimes it works (reusable rockets, Starlink). Sometimes it does not (the Cybertruck’s production timeline, Robotaxis by 2020).

We have been here before with Tesla’s compute ambitions. Back in October, we analyzed Musk’s proposal to turn Tesla’s parked fleet into a distributed AI network, essentially using idle vehicles as inference processors. The concept faced the same fundamental problem: solving AI compute constraints through unconventional infrastructure rather than conventional scale.

EVXL’s Take

This is the third major Dojo pivot we have documented, and the pattern is becoming clear. Tesla’s custom silicon ambitions exist primarily as narrative rather than production reality.

The company committed $500 million to Buffalo for Dojo infrastructure, then killed it without informing state officials. It positioned Dojo as central to achieving Full Self-Driving, then disbanded the team and called it an evolutionary dead end. Now it is resurrecting the project for an entirely different purpose that conveniently aligns with Musk’s other company’s IPO strategy.

Meanwhile, Nvidia just open-sourced the foundation for reasoning-based autonomous driving and handed it to every automaker in the world. Mercedes is shipping it this quarter. Tesla’s response? Talking about satellites.

The space data center concept may eventually prove transformative. But for Tesla investors focused on the company’s core EV and autonomy business, Dojo3 is a distraction at best and a misdirection at worst. Expect the next Tesla earnings call to feature more questions about when Dojo3 satellites will launch than about why FSD still requires driver supervision seven years after it was sold as a feature.

Editorial Note: This article was researched and drafted with the assistance of AI to ensure technical accuracy and archive retrieval. All insights, industry analysis, and perspectives were provided exclusively by Haye Kesteloo and our other EVXL authors, editors, and Youtube partners to ensure the “Human-First” perspective our readers expect.