Tesla Completes AI5 Chip Design Review, Targets 2026 Mass Production with TSMC N3P Process

 Tesla announced that it has completed the design review of its next-generation AI5 inference chip, which CEO Elon Musk described as an “epic-level product.” The chip is built on TSMC’s advanced N3P process and is scheduled for initial deployment in the company’s Optimus humanoid robot by December 2024, with mass production expected to ramp up in 2026.

Musk emphasized that while AI5 may not be the best automotive AI chip, it is designed to deliver industry-leading performance for AI inference models with up to 250 billion parameters, offering superior cost efficiency and power-to-performance ratio compared with existing products. The follow-on AI6 chip, also under development, is expected to achieve an even greater leap in performance and could become Tesla’s most advanced AI processor to date.

In August, Tesla disbanded its Dojo supercomputer team, shifting its R&D focus away from self-developed training chips toward inference-optimized chips. “All of our engineering talent will now be focused on AI5, AI6, and future inference-oriented chips,” Musk said. He added that consolidating efforts onto a single architecture will accelerate development and enhance performance across Tesla’s product lines.

The AI5 chip will initially be fabricated at TSMC’s facilities in Taiwan, followed by expanded production at its new Arizona plant in the United States by late 2026. In contrast, the AI6 chip will be manufactured by Samsung at its new Texas facility. According to Musk, AI6 is designed to combine both inference and training capabilities, potentially replacing the Dojo platform for future large-scale AI model training.

These new chips are pivotal to Tesla’s upcoming products, including the Optimus humanoid robot and the Cybercab autonomous taxi, as the company continues to strengthen its in-house semiconductor strategy to support AI-driven innovation.

Source: China Times