EVwire brief: Elon Musk announced that Tesla has taped out its next-generation AI5 chip and shared the first real-world photo of the new hardware.
Musk said AI5 could become “one of the most produced AI chips ever,” with manufacturing support from TSMC and Samsung. He also confirmed that AI6, Dojo3, and “other exciting chips” are in development.
Context:
AI5 is Tesla’s custom System-on-Chip (SoC) designed for real-time AI inference in vehicles and robots.
Compared to AI4, the chip is expected to deliver 8–10x raw compute, ~9x more memory capacity (from ~16GB in AI4), and ~5x improved memory bandwidth, with Musk stating that a single AI5 has ~5 times the useful compute of a dual SoC AI4 setup.
Power consumption of the AI5 system is estimated at up to ~800 watts under heavy load (with efforts to bring it closer to ~250W in optimized scenarios), compared to ~160–200 watts for AI4 and ~100 watts for HW3.
Tesla designs its own chips to work optimally with its own products
Tesla has indicated that performance per watt remains a key focus, with the chip optimized specifically for neural network workloads.
AI5 is expected to be manufactured using advanced process nodes in the 2nm–3nm class, with production split between TSMC and Samsung. Initial production is targeted for late 2026 to early 2027, with volume ramp expected in 2027.
Deployment in vehicles is expected to follow, with no retrofit path for older HW3 or HW4 systems, likely due to architectural differences.
AI5 will be deployed to products like Optimus
AI5 scales Tesla’s AI hardware for vehicles and robotics
AI5 is designed to power Tesla’s Full Self-Driving system and Optimus humanoid robot, acting as the onboard inference engine for real-time decision-making. Elon Musk also noted that the chip will be used for Tesla’s supercomputer clusters.
A single AI5 chip is expected to deliver performance comparable to an Nvidia H100 GPU for inference workloads, while a dual-chip configuration approaches the performance of Nvidia’s B100/B200 class, as per Musk.
Elon discussed Tesla’s AI chip program in this video from the 2025 Tesla Annual Shareholder Meeting:
The chip removes traditional components such as a dedicated GPU and image signal processor, consolidating workloads into a unified neural network architecture. This allows for higher efficiency and better silicon utilization.
Tesla is also targeting high-volume production, with AI5 expected to be deployed across millions of vehicles and robots.
Small-batch production is planned for 2026, with scaling supported by external foundries and future internal manufacturing initiatives such as Tesla’s Advanced Technology Fabrication facility near Giga Texas.
Source: Elon Musk on X
DIG DEEPER into the Tesla industry news with our dedicated TESLAWIRE page. And don’t forget to subscribe to our EV industry newsletter to join 14,000+ EV geeks.
