Tesla AI5 tape‑out

A tape-out is the moment a chip design is frozen and sent to a factory, and Tesla says its next self-driving processor, AI5, has reached that step. (electrek.co) Elon Musk posted on April 15 that Tesla’s chip team had taped out AI5 and said AI6 and Dojo3 were also in development. Electrek reported the milestone still leaves volume production more than a year away. (electrek.co) Tesla builds these chips for inference, the part of artificial intelligence that runs trained models inside a car or robot in real time, rather than training giant models in a data center. Tesla says on its AI page that it is pursuing “efficient use of inference hardware” for vehicles, robots and related systems. (tesla.com) Musk told Tesla investors in October 2025 that AI5 would be manufactured by Samsung in Texas as well as Taiwan Semiconductor Manufacturing Co. in Arizona. He said Tesla wanted “excess production,” with any chips not used in cars or robots available for data centers. (cnbc.com) That manufacturing split put Tesla’s chip plan inside a broader U.S. push to localize advanced semiconductor capacity. TSMC said in March 2025 that it was expanding its U.S. investment to $165 billion, and Samsung said its Taylor, Texas fab is part of a multibillion-dollar U.S. buildout backed by up to $6.4 billion in CHIPS Act funding. (tsmc.com) (samsung.com) The timing also shows how far Tesla’s silicon schedule has slipped. Electrek reported on April 15 that the tape-out came nearly two years after Tesla had originally said AI5 would already be in vehicles. (electrek.co) Musk had said in January 2026 that the AI5 design was “almost done,” months after earlier saying it was finished. By November 2025, he had said volume availability would not arrive until mid-2027, with Cybercab launching on AI4 hardware instead. (electrek.co 1) (electrek.co 2) Tesla is pitching AI5 against Nvidia-style performance, but the company is not trying to replace Nvidia across the board. CNBC reported in October 2025 that Musk said Tesla was “not about to replace Nvidia” even as it expanded its own chip program. (cnbc.com) That distinction matters in chip terms: Nvidia’s H100 is a data-center training and inference processor, while Tesla’s in-house silicon is designed to fit the power, heat and cost limits of a vehicle or robot. Tesla’s own AI materials frame the problem as deploying autonomy at scale with specialized inference hardware. (tesla.com) (cnbc.com) The next test is no longer design but manufacturing. If AI5 moves from tape-out to stable production on the schedule Musk outlined, Tesla’s custom chip effort will be measured in cars and robots on the road, not in posts about what comes after AI4. (electrek.co) (cnbc.com)