Tesla Pulls the Plug: The End of Dojo and a Pivot in the Self-Driving Race

In 2019, the world watched as Elon Musk unveiled Project Dojo: a new breed of AI supercomputer, powered by custom Tesla-designed D1 chips, that promised to devour mountains of driving video and train neural networks poised to revolutionize autonomous vehicles. Dojo’s technical ambitions dazzled more than an exaflop of power, a dense fabric of interconnected processing units, and the hope of rendering the supply constraints and architectural limits of external chips irrelevant.

Wall Street and tech forums alike caught the fever: analysts speculated Dojo could add hundreds of billions to Tesla’s worth thanks to its first-mover lead in self-driving and the robotaxi revolution. It was Musk at his storytelling best. Dojo would be the digital cortex underpinning a fleet of Tesla cars learning and improving together on the open road.

The Reality Check: Talent and Tech Turbulence

By mid-2025, the dream began to unravel. Key leaders, including Dojo’s own head, Peter Bannon, departed. A mass exodus saw around 20 top engineers bolt to form Density AI, a stealth startup helmed by former Dojo chief Ganesh Venkataramanan. Internal momentum faltered just as cracks began to show in Tesla’s robotaxi rollout: limited launches, driving hiccups, and investor scepticism.


Elon Musk, ever pragmatic in the face of market tides, changed tack. According to Musk, maintaining two divergent chip lines. Dojo’s custom training chips and Tesla’s new inference-focused AI5 and AI6 processors no longer made business sense. Dojo’s mission, powering FSD (Full Self-Driving) through brute-force on in-house hardware, lost out to a strategy leveraging external chip partnerships with Nvidia, AMD, and Samsung.

The Shift: From “Build” to “Buy”

The end of Project Dojo doesn’t mark a retreat from autonomy or AI, just a strategic pivot. Tesla has inked a $16.5 billion deal with Samsung to supply next-generation AI chips for its vehicles and humanoid robots, a clear sign that it believes the path forward runs faster on Nvidia’s and Samsung’s silicon than on furrowed brows in Tesla’s own labs.

Musk confirmed on X (formerly Twitter) that the company will now focus exclusively on AI5 and AI6 chips, which, while not exclusively designed for training, will be “excellent for inference and at least pretty good for training.” The plan: integrate these chips into Tesla products like the Optimus robot and Cybercab while outsourcing vast data crunching and AI training horsepower to industry partners.

Tesla’s Dojo once symbolized a Silicon Valley ethos: solve the world’s hardest problems with audacity and your own hardware. Its end is a reminder that, sometimes, winning the future means knowing when to build—and when to buy.