On January 18th, Musk took to social media platform X to declare: "With the AI5 chip design finalized, we're reviving the Dojo 3 project." This succinct announcement sent ripples across the tech sphere.

This move comes a mere five months after Tesla's abrupt decision in August 2025 to suspend the Dojo project and disband its core team.

Back in August 2025 when the Dojo project was put on hold, Musk openly stated that it was illogical to pursue the development of two vastly different AI chip designs simultaneously.

The conventional Dojo system is built on Tesla's proprietary D1 chip and intricate wafer-level packaging technology. In contrast, Dojo 3 embraces a revolutionary multi-chip integration design. Musk explained that this approach entails assembling multiple chips onto a single motherboard.

Such a design can substantially streamline the network architecture and slash the overall costs of supercomputer clusters. Musk dubbed this integrated solution "Dojo 3."

In tandem with the Dojo 3 revival announcement, Musk also unveiled a unique job posting, extending an invitation to engineers eager to be part of crafting the world's highest-volume chip.

The Tesla CEO specifically asked applicants to distill the most formidable technical challenge they've overcome into three key points.

Tesla has embraced a forward-thinking dual-foundry strategy for its AI chips. The AI5 chip will be jointly manufactured by TSMC and Samsung, albeit with slightly varying processes between the two.

Samsung Electronics' iteration of the AI5 chip will be crafted using a 2-nanometer process, sharing the same process node as the upcoming AI6 chip. To cater to Tesla's urgent need for the AI5 chip, Samsung has planned to bring forward the production start date of its Taylor, Texas, facility from the originally slated end of 2026 to the third quarter of that year.

According to Musk's outlined plan, AI5 chip samples and limited production are projected for 2026, with mass production not expected until 2027. The AI6 chip is slated to enter mass production in mid-2028, with an anticipated performance boost of roughly double.

The driving force behind Tesla's swift resurrection of the Dojo 3 project is the escalating computational demand from its full self-driving, humanoid robot, and Robotaxi ventures.

In August 2025, a Tesla subsidiary secured a ride-hailing license in Texas, paving the way for autonomous taxi services across the state, with vehicles even permitted to operate sans human safety drivers.

In Austin, Texas, Tesla has already rolled out a limited autonomous ride-hailing service. Tesla's FSD has amassed nearly 10 billion kilometers of driving experience.

According to a CSC Securities report, by the first quarter of 2026, Tesla aims to finalize the design of the Optimus Gen3 humanoid robot and commence mass production in the second half of the year, targeting an annual output of 50,000 units.

Since its launch in July 2023, the Dojo supercomputer has been primarily tasked with processing approximately 160 billion frames of video data generated daily by Tesla's global fleet. Leveraging unsupervised learning algorithms, it continuously refines the neural network training for the full self-driving system and the Optimus humanoid robot.

References:

https://finance.eastmoney.com/a/202601193623016759.html

https://stockfinance.sina.cn/stock/go.php/paper/reportid/820329720242/index.phtml?vt=4

https://www.ithome.com/0/894/327.htm