Tesla has disbanded its Dojo supercomputer team, marking a significant strategic shift toward relying on external AI training resources and focusing on vehicle-integrated chips.
At a Glance
- Tesla disbanded Dojo team; leader Peter Bannon is leaving
- Engineers reassigned to other compute projects inside Tesla
- Musk shifts to in-vehicle “AI6” chips and outsources large-scale training
- Around 20 former Dojo engineers moved to startup DensityAI earlier in 2025
- Nvidia and AMD gain as key external AI training hardware providers
From Bespoke Supercomputer to External AI Training
Tesla’s Dojo project, designed to develop a proprietary supercomputer optimized for training autonomous driving and robotics AI models, has been officially disbanded. The effort, once viewed as a strategic advantage to reduce dependence on Nvidia GPUs, faltered amid strong competition from established AI hardware vendors. Bloomberg reports that Dojo leader Peter Bannon is exiting Tesla, and remaining team members have been reassigned internally.
Previously, Dojo aimed to accelerate Tesla’s AI training with custom D1 chips tailored to the company’s extensive video data, promising a unique edge in machine learning workloads. However, as Nvidia and AMD’s AI accelerators demonstrated faster scalability and more mature ecosystems, Tesla’s bespoke approach struggled to keep pace. The departure of roughly 20 Dojo engineers earlier this year to DensityAI, a startup founded by former Tesla personnel, signaled internal challenges in retaining talent and sustaining momentum.
Watch now: Tesla Disbands Dojo Supercomputer Team, Upending AI Effort – Bloomberg · YouTube
Musk’s Pivot to AI6 Chips and Outsourcing
Elon Musk framed the shift as a pragmatic focus on next-generation vehicle chips, termed “AI6,” optimized primarily for AI inference inside Tesla’s cars and robots. Meanwhile, Tesla will rely more heavily on third-party cloud providers such as Amazon Web Services and Oracle Cloud to handle the computationally intensive training workloads. This model leverages mature, scalable external infrastructure while maintaining Tesla’s core differentiation in integrated vehicle hardware and software.
This transition could streamline Tesla’s development cycles for features like Full Self-Driving and Optimus robots by reducing capital investment in custom data centers. However, it also introduces dependencies on external cloud vendors, which may impact cost, availability, and proprietary advantage. Industry analysts caution that the shift narrows Tesla’s in-house AI hardware moat but could accelerate product deployment and cost control.
Market Impact and Supply Chain Effects
The strategic pivot benefits AI hardware leaders Nvidia and AMD, whose accelerators will supply Tesla’s training needs. Samsung remains a key manufacturing partner for Tesla’s AI6 chips. Meanwhile, DensityAI is positioned to capitalize on talent migrating from Tesla, reflecting broader fragmentation in the AI hardware startup landscape.
Investor perspectives on Tesla’s AI competitiveness are recalibrating, weighing the loss of a bespoke training stack against gains from a more focused vehicle inference roadmap and external training scalability. Technology commentators note some debate over whether Dojo is completely “dead” or evolving into new internal AI initiatives. Nonetheless, Bloomberg’s reporting and Musk’s statements provide strong evidence of the major organizational and strategic redirection.
Sources
