Tesla is already planning the next generation of its FSD computer hardware – even though the AI5 chip is not yet on the market. The upcoming AI6 chip is not only expected to be significantly more powerful but also to bring a fundamental architecture change. It will be manufactured by Samsung in a new semiconductor factory in Texas – with direct support from Tesla for production optimization.
The current problem: Two languages for one system
Currently, Tesla trains its FSD models on the supercomputer cluster Cortex, which is based on Nvidia's H200 GPUs. However, the vehicles themselves use Tesla's own chips (HW3/AI3 or HW4/AI4).
This means: Every new improvement must first be developed and validated on Nvidia hardware before it is adapted for Tesla's own chips and tested again.
Although Tesla already uses AI4 chips in the training systems in parallel, the intermediate step remains – and slows down the process from training to rollout of new features.
AI6: Unified architecture for training and vehicle
With AI6, Tesla wants to completely solve this problem. The chip is directly based on the Dojo supercomputer architecture, which will be used both in the data center for training and in the vehicle and the humanoid robot Optimus in the future.
This means there will be only one hardware platform in the future – the complex translation process is eliminated. The result: faster development cycles, shorter update intervals, and a significantly higher pace of innovation.
Manufacturing partnership with Samsung
The manufacturing of AI6 is taken over by Samsung in a new fab in Texas. The special thing is: Tesla is directly involved in the production process to achieve maximum efficiency – an unusually close exchange between chip designer and manufacturer.
Elon Musk announced that he would personally inspect the production lines to accelerate progress. The goal is to combine Tesla's know-how in automation and production optimization with Samsung's manufacturing expertise.
Why AI6 is so important
While AI5 will already bring a significant performance increase over AI4, AI6 is a real technological leap:
-
Unified hardware for training and use in the vehicle
-
Elimination of the “Two Languages” issue between Nvidia training and Tesla hardware
-
Faster iterations on FSD updates
-
Synergies with Optimus and other AI projects
Conclusion: Tesla is building a self-reinforcing AI ecosystem
With AI6, Tesla lays the foundation for a completely integrated AI infrastructure – from training to hardware development to vehicle and robotics integration. This is likely to not only accelerate the development of Full Self-Driving, but also advance future applications in robotaxis, humanoid robots, and industrial automation.