Tesla shuts down Dojo supercomputer project in August 2025

Tesla announced the shutdown of its Dojo supercomputer and the disbanding of the team behind it in mid‑August 2025, a move that follows a $16.5 billion agreement with Samsung for next‑generation AI6 chips and marks a strategic pivot away from in‑house AI hardware [1].

End of Dojo, rise of AI6

Tesla’s decision to close Dojo came just weeks after securing a massive $16.5 billion contract with Samsung for AI6 system‑on‑chip processors. Elon Musk described Dojo 2 as “an evolutionary dead end” and indicated that the new AI6 chips would fulfill the compute needs that Dojo was meant to address [1]. The shutdown also saw the exit of Dojo lead Peter Bannon and roughly 20 engineers who later founded the AI chip startup DensityAI [1]. Bloomberg reported the project’s disbandment in August 2025, though a later note indicated a restart in January 2026 [2].

Impact on Tesla’s AI roadmap

Dojo was originally built to train Tesla’s Full Self‑Driving (FSD) neural networks on millions of terabytes of video data collected from its 4 + million cars [2]. Its goal of scalable video processing positioned it as a cornerstone of Tesla’s vision‑only autonomous‑driving strategy. With Dojo’s closure, Tesla will rely on the AI6 chips—designed to scale from vehicle‑level FSD to data‑center‑grade training—to continue model development. Analysts see the loss of specialized talent as a risk to the project’s continuity, but the partnership with Samsung may accelerate hardware availability compared with the in‑house development timeline [1].

Competitive context

Tesla’s primary GPU cluster, built on 5,760 Nvidia A100 GPUs, was once touted as roughly the world’s fifth‑largest supercomputer at about 81.6 petaflops [2]. By contrast, Dojo’s custom D1 chips aimed to reduce latency and increase bandwidth, a goal now transferred to Samsung’s AI6 silicon. Competitors such as Nvidia continue to dominate the high‑performance AI training market, and Tesla’s shift to external chip sourcing aligns it more closely with industry standards rather than maintaining a unique hardware stack.

What to watch

• AI6 chip rollout – monitor Samsung’s production timeline and integration into Tesla’s vehicles and data‑center training clusters.
• Dojo restart – watch for any updates on the reported January 2026 restart and whether the new effort will differ from the original architecture.
• FSD performance metrics – track any changes in FSD accuracy or robotaxi deployment that can be linked to the new hardware platform.

The shutdown underscores Tesla’s willingness to abandon costly internal projects when external solutions appear more viable, raising questions about the company’s long‑term hardware independence and its ability to sustain rapid AI advances without a dedicated supercomputing asset.