Podcast Episode
However, on 19 January 2026, Musk announced on X that Tesla would resume work on Dojo3 following progress on the AI5 chip. He wrote that now the AI5 chip design is in good shape, Tesla will restart work on Dojo3, and invited job applications for what he described as the highest volume chips in the world.
Expected performance is between 2,000 and 2,500 TOPS, trillion operations per second, putting it in competitive territory with high-end graphics processors whilst consuming significantly less power. Musk has claimed the AI5 offers 10 times the performance per dollar compared to NVIDIA AI chips, a crucial factor in Tesla's decision to develop custom silicon rather than rely solely on off-the-shelf solutions.
The chip is being manufactured through partnerships with both TSMC and Samsung, with production facilities in Arizona and Texas respectively. Volume production is targeted for late 2027, though a small batch is expected in late 2026.
The advantage of a custom supercomputer lies in its optimisation for Tesla's specific workloads. Whilst general-purpose systems from NVIDIA and other manufacturers offer flexibility, they include capabilities Tesla does not need whilst potentially lacking optimisations for video processing and neural network training specific to automotive applications.
Analysts have dubbed 2026 a prove-it year for Tesla's self-driving technology. The company has repeatedly missed deadlines for delivering full autonomy, and competitors including Waymo, Cruise, and traditional manufacturers are making significant progress in the autonomous vehicle space.
The AI5 chip's first application will be in the Cybercab, where the enhanced computing power will support additional safety and redundancy systems. Only after proving itself in the robotaxi will AI5 be introduced into Tesla's consumer vehicle lineup, including Model 3, Model Y, Model S, and Model X.
This timeline is exceptionally aggressive compared to typical semiconductor development cycles, which often span multiple years between major releases. However, the rapid iteration could give Tesla a competitive advantage if the company can maintain quality whilst increasing development velocity.
The financial implications are significant. Training costs for advanced AI systems can run into hundreds of millions of pounds, and reducing these expenses whilst improving performance could fundamentally alter the economics of autonomous driving development.
The company is also competing for engineering talent in an extremely competitive market. Musk's call for top engineers to join the project suggests Tesla recognises it needs significant expertise to succeed where the previous attempt failed.
Manufacturing at scale presents another challenge. Whilst TSMC and Samsung are proven partners, ramping production to the volumes Tesla requires for both its data centres and vehicle fleet will test the supply chain.
The success or failure of Dojo3 and the AI5 chip will likely have profound implications not just for Tesla but for the broader autonomous vehicle industry. If Tesla can achieve the performance and cost advantages it claims, other manufacturers may need to reconsider their reliance on commercial chip suppliers. Conversely, another failed attempt could validate the approach of competitors who have chosen to work with established AI hardware providers rather than developing custom solutions.
Tesla Revives Dojo3 Supercomputer Project After Dramatic U-Turn on AI Strategy
January 19, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
Tesla has announced it will restart development of its Dojo3 supercomputer project, just five months after CEO Elon Musk declared the previous generation an evolutionary dead end and disbanded the team. The reversal comes as Tesla makes significant progress on its next-generation AI5 chip, which will form the backbone of the revived project.
The Dramatic Reversal
In August 2025, Tesla shut down its Dojo supercomputer team and dismissed Peter Bannon, the project lead who had been instrumental in developing the system. At the time, Musk described Dojo2 as an evolutionary dead end and stated it no longer made sense for Tesla to divide its resources and scale two quite different AI chip designs. The decision appeared to mark the end of Tesla's ambitious in-house supercomputer programme.However, on 19 January 2026, Musk announced on X that Tesla would resume work on Dojo3 following progress on the AI5 chip. He wrote that now the AI5 chip design is in good shape, Tesla will restart work on Dojo3, and invited job applications for what he described as the highest volume chips in the world.
The AI5 Chip at the Core
The AI5 chip represents a substantial leap in performance over Tesla's current AI4 hardware. The new chip is expected to deliver up to 50 times improvement in certain performance metrics, with specifications including 40 times faster processing speed, 8 times the raw computing power, 9 times the memory capacity, 5 times the memory bandwidth, and 3 times the efficiency per watt compared to AI4.Expected performance is between 2,000 and 2,500 TOPS, trillion operations per second, putting it in competitive territory with high-end graphics processors whilst consuming significantly less power. Musk has claimed the AI5 offers 10 times the performance per dollar compared to NVIDIA AI chips, a crucial factor in Tesla's decision to develop custom silicon rather than rely solely on off-the-shelf solutions.
The chip is being manufactured through partnerships with both TSMC and Samsung, with production facilities in Arizona and Texas respectively. Volume production is targeted for late 2027, though a small batch is expected in late 2026.
The Original Dojo Project
Tesla first deployed its Dojo supercomputer in 2023 as a custom-built training system for the company's AI software, particularly its Full Self-Driving technology. The system was designed to process massive volumes of video data collected from Tesla vehicles operating around the world, using this real-world information to train increasingly sophisticated autonomous driving models.The advantage of a custom supercomputer lies in its optimisation for Tesla's specific workloads. Whilst general-purpose systems from NVIDIA and other manufacturers offer flexibility, they include capabilities Tesla does not need whilst potentially lacking optimisations for video processing and neural network training specific to automotive applications.
Implications for Autonomous Driving
The restart of Dojo3 comes at a critical juncture for Tesla's autonomous vehicle ambitions. The company is preparing to launch its Cybercab robotaxi, scheduled for volume production in April 2026. This vehicle, designed without traditional driver controls, represents Tesla's vision for fully autonomous urban transport and requires exceptional AI performance and reliability.Analysts have dubbed 2026 a prove-it year for Tesla's self-driving technology. The company has repeatedly missed deadlines for delivering full autonomy, and competitors including Waymo, Cruise, and traditional manufacturers are making significant progress in the autonomous vehicle space.
The AI5 chip's first application will be in the Cybercab, where the enhanced computing power will support additional safety and redundancy systems. Only after proving itself in the robotaxi will AI5 be introduced into Tesla's consumer vehicle lineup, including Model 3, Model Y, Model S, and Model X.
Aggressive Development Roadmap
Alongside the AI5 announcement, Musk revealed that work on AI6 has already entered early development stages. The company is targeting an ambitious nine-month design cycle for future chip generations, with AI7, AI8, and AI9 already on the roadmap.This timeline is exceptionally aggressive compared to typical semiconductor development cycles, which often span multiple years between major releases. However, the rapid iteration could give Tesla a competitive advantage if the company can maintain quality whilst increasing development velocity.
The Broader Strategy
Tesla's approach represents a hybrid model for AI infrastructure. The company does not intend to replace NVIDIA data centre GPUs entirely but aims to complement them with custom silicon optimised for specific workloads. This strategy mirrors approaches taken by major technology companies including Google, Amazon, and Meta, all of which have developed custom AI accelerators whilst continuing to use commercial offerings.The financial implications are significant. Training costs for advanced AI systems can run into hundreds of millions of pounds, and reducing these expenses whilst improving performance could fundamentally alter the economics of autonomous driving development.
Questions and Challenges
Significant questions remain about whether Tesla can avoid the pitfalls that led to Dojo2 being abandoned. The reasons for the previous generation's failure have not been publicly detailed, raising concerns about whether similar issues might affect Dojo3.The company is also competing for engineering talent in an extremely competitive market. Musk's call for top engineers to join the project suggests Tesla recognises it needs significant expertise to succeed where the previous attempt failed.
Manufacturing at scale presents another challenge. Whilst TSMC and Samsung are proven partners, ramping production to the volumes Tesla requires for both its data centres and vehicle fleet will test the supply chain.
The success or failure of Dojo3 and the AI5 chip will likely have profound implications not just for Tesla but for the broader autonomous vehicle industry. If Tesla can achieve the performance and cost advantages it claims, other manufacturers may need to reconsider their reliance on commercial chip suppliers. Conversely, another failed attempt could validate the approach of competitors who have chosen to work with established AI hardware providers rather than developing custom solutions.
Published January 19, 2026 at 12:50pm
