On September 7th, Tesla CEO Elon Musk revealed a major chip R&D breakthrough on X: after reviewing the AI5 chip with its team, he called it "epic," while teasing the upcoming AI6 as the "best AI chip ever"—grabbing instant industry attention.
Musk noted the AI5 will likely top all inference chips for models under 25 billion parameters, boasting the lowest silicon cost and industry-leading performance-per-watt ratio—key strengths for low-to-mid parameter model inference.
Tesla has also made a critical R&D shift: moving from dual parallel chip architectures to a single focus. Musk stressed this concentrates talent and cuts resource waste, adding (as he did in August) that future chips (AI5, AI6, etc.) must handle both inference and training to be Tesla’s AI ecosystem core.
Tech & production details:
- AI5: Uses 3nm N3P process, 2000-2500 TOPS (4-5x AI4’s power), supports unsupervised learning. Focused on autonomous driving inference/cluster training; mass production via TSMC in late 2026 (Samsung later).
- AI6: Touted as Tesla’s AI ecosystem "unified heart." First samples from Samsung’s S. Korea factory, later shifted to Samsung’s Texas (U.S.) plant (2025 launch). Initially for Cybercab/Optimus, then AI data centers—targeting NVIDIA’s H200 GPU.
Musk also recruited R&D talents for "life-saving chips." The industry sees this shift/breakthrough as Tesla’s full push into self-developed hardware, using core chips to power autonomous driving/robots and speed up its AI ecosystem.
ICgoodFind concludes: Tesla’s AI5/AI6 progress shows automakers’ ambition in high-end chips. Their process wins and scenario focus may reshape the semiconductor industry’s competitive landscape.
