Tesla Advances In-House AI Chips, AI5 Design Reviewed

Recently, Tesla has achieved a major milestone in its in-house chip development. On September 7, CEO Elon Musk announced via social media that the design review for the AI5 chip is complete, describing it as a "legendary" product. Musk also hinted that the upcoming AI6 chip could become the most advanced AI chip the company has ever created.

Tesla had previously disclosed that it was internally testing the integration of thousands of AI5 chips for next-generation AI model training, and early results have been promising. According to Musk, the AI5 chip may offer the best inference performance for models with fewer than 250 billion parameters, delivering top efficiency while keeping silicon costs low. The AI6 chip is expected to build on this foundation with even higher performance.

Notably, Tesla halted its Dojo chip project in August, with project lead Peter Bannon departing the company. Musk explained that splitting resources across two different AI chip designs was no longer practical. Previously, Tesla had pursued a dual-track approach—Dojo chips for training and FSD chips for in-vehicle inference—but the company is now focusing on a single architecture to concentrate talent and deliver superior products.

The AI5 and AI6 chips are designed to support Tesla's AI and autonomous driving initiatives. The AI5 chip, a transitional or scenario-specific workhorse, will be fabricated by TSMC and is expected to enter mass production by the end of 2026. The AI6 chip will act as the "central brain" of Tesla's future AI ecosystem, manufactured by Samsung Electronics, initially powering Tesla's Cybercab robotaxi and Optimus humanoid robots, with potential expansion to AI data centers. First AI6 samples will be produced at Samsung's Korean facility, with mass production planned at Samsung's Texas fab, scheduled to start operations in 2025, positioning Tesla to compete with industry leaders such as Nvidia's H200 GPU.