Tesla Tapes Out AI5 Chip
Elon Musk announced that Tesla has taped out its AI5 chip and shared hardware photos while confirming AI6, Dojo3 and other chips are in development. The social posts showing the AI5 tape-out received substantial engagement and highlighted Tesla's continued in-house AI hardware push. (x.com) (x.com)
Tesla has taped out AI5, the next chip for the computers that run its driver-assistance and robotics software, Elon Musk said on April 15. (electrek.co) In chipmaking, a tape-out means the design is finished and sent to a factory for fabrication; it does not mean the chip is already in cars. Musk said AI6, Dojo3 and other chips are also in development, and he posted photos of the AI5 hardware on X. (electrek.co) (ibtimes.com.au) Tesla says its in-car AI computer is built to process neural networks locally, and its Full Self-Driving (Supervised) system uses exterior cameras and over-the-air software updates. On Tesla’s support pages, the company says current vehicles still require active driver supervision and are not autonomous today. (tesla.com 1) (tesla.com 2) That distinction matters because Tesla is trying to build more of the computing stack itself as it pushes robotaxis, Optimus robots and larger driving models. In its 2025 annual update, Tesla said it was shifting from a “hardware-centric business” to a “physical AI company” and investing in AI silicon and training infrastructure. (assets-ir.tesla.com) Tesla’s current public product pages center on Full Self-Driving (Supervised), which is available in eight countries and has logged more than 9.26 billion miles, according to the company. Tesla also says its fleet now exceeds six million vehicles, giving it a large stream of driving video for training. (tesla.com) Musk’s post does not put AI5 into production immediately. Electrek reported that automotive-grade validation and ramp-up after tape-out typically take 12 to 18 months, and that Tesla had previously moved its AI5 vehicle timeline more than once. (electrek.co) That delay has practical consequences inside Tesla’s lineup. Electrek reported that the Cybercab is still set to launch on AI4 hardware, and Tesla’s support pages show the company is still upgrading some older customers to Full Self-Driving computer 3.0 rather than a newer generation. (electrek.co) (tesla.com) Tesla has been building custom AI chips for years, not just for cars but also for model training. At AI Day in 2021, the company introduced the Dojo D1 chip and a training tile, a server building block for the Dojo supercomputer used to train vision models. (datacenterdynamics.com) Musk said AI5 “will be one of most produced AI chips ever” and thanked Taiwan Semiconductor Manufacturing Company and Samsung for support, according to reports quoting his follow-up post. The immediate takeaway is narrower: Tesla has locked the design for its next chip, but the harder part now is turning that design into tested hardware at automotive scale. (ibtimes.com.au) (electrek.co)