Tesla’s chip and Dojo supercomputer chief departs company

Tesla is parting ways with one of its most influential hardware leaders. Pete Bannon, the company’s chip and Dojo supercomputer chief, has departed after nearly a decade of service. 

He joined Tesla in 2016 from Apple, where he had been a key figure in designing the company’s A-series chips. At Tesla, Bannon became instrumental in building the company’s custom silicon capabilities, overseeing projects to help define the automaker’s AI ambitions.

Bannon’s most high-profile responsibility was leading the development of Dojo, Tesla’s in-house supercomputer designed specifically for artificial intelligence training. He reported directly to CEO Elon Musk and was considered central to Tesla’s strategy to reduce reliance on external AI hardware providers like Nvidia. 

His departure comes as Tesla makes a significant strategic pivot. The company has decided to shut down the Dojo program entirely, dissolving the team that has been working on it for years. Engineers from the project are being reassigned to other computing and data center initiatives within Tesla’s broader operations. Bloomberg was the first to break the news, later confirmed by multiple industry sources, that Musk personally ordered the change in direction.

At Tesla, Dojo stood out as one of the company’s most ambitious technical projects. The system was intended to be a bespoke AI training platform driven by Tesla-designed chips that could handle the vast amount of video and sensor data collected daily from its global fleet. That data is essential to advancing Tesla’s Full Self-Driving (FSD) and robotics programs, for which Dojo was expected to be central.

Tesla winding down the project represents a huge departure from its AI roadmap. The move suggests the company is stepping back from a strategy of building its own AI chips effectively from scratch toward working more closely with established chip and hardware makers. 

Musk orders reboot of AI strategy

Elon Musk is changing Tesla’s AI strategy. Instead of developing its own computing hardware, Tesla will outsource more to vendors like Nvidia, AMD, and Samsung.

The deal signed with Samsung by Tesla for its future-gen A16 AI chips is estimated at $16.5 billion. The company is also rapidly ramping up using Nvidia’s AI training high-performance GPUs.

The move is a significant departure from Musk’s initial vision for Dojo, which was intended to become the AI equivalent of Tesla’s well-regarded Supercharger network and a point of differentiation that would help it outpace rivals. 

The Dojo project has faced technical challenges and delays since its unveiling in 2021.

Many engineers who worked on the system have left the company. Around 20 have already moved on to launch an AI startup called DensityAI.

Tesla’s decision to close the Dojo team effectively ends its in-house supercomputing push. The company will now channel resources into other AI infrastructure and driver-assistance projects.

On the company’s July earnings call, Musk had said Tesla expected the latest Dojo version to run at scale next year. But the abrupt shutdown suggests priorities have shifted.

Tesla eyes partnerships for compute needs

The service marks a shift for Tesla, which, up until now, has been doing all training computing internally. Most of the heavy lifting will be done by Nvidia’s latest GPUs. AMD is also likely to contribute processing for some projects.

Samsung will play a key role in Tesla’s chip production. It’s a way for Tesla to tap into state-of-the-art hardware at little risk and cost instead of just building its gargantuan supercomputers from scratch. 

The move comes as Tesla tests limited Robotaxi services in Austin and San Francisco. In Austin, a human safety supervisor still rides in the passenger seat. In San Francisco, human drivers operate the vehicles while users hail them through a “Tesla Robotaxi” app.

Tesla has seen an exodus of senior talent this year, with the departure of Bannon adding to the list. Former Optimus robotics head Milan Kovac, software VP David Lau, and Musk’s former chief of staff Omead Afshar have also left.

Get seen where it counts. Advertise in Cryptopolitan Research and reach crypto’s sharpest investors and builders.