Elon Musk announced on Wednesday that Tesla's chip design team has completed the tape-out of the AI5 processor, the company's next-generation AI chip. He showed the first physical sample on X and thanked "TSC" for their production support. He meant TSMC. Both TSMC and Samsung will manufacture the chip.

Tape-out means the design is finalized and sent to the foundry for fabrication. It is a genuine milestone. But context matters: the AI5 was originally expected to tape out in late 2024. It is roughly two years behind schedule. Musk claims 40x performance over the AI4 predecessor, a number that drew immediate skepticism from industry analysts.

Tom's Hardware noted the chip sample carries a "KR 2613" marking, suggesting it was packaged in the 13th week of 2026, roughly late March. That tracks with a tape-out in Q1 2026. Full-scale mass production is expected in 2027, about a year after tape-out completion, which is standard for custom silicon at this scale.

The Real Target: Optimus and Cybercab

The AI5 is not a data center chip competing with Nvidia. It is an edge AI processor designed for Tesla's products: autonomous vehicles, the Cybercab robotaxi, and the Optimus humanoid robot. Tesla needs custom silicon because running inference on general-purpose GPUs at the edge is too power-hungry and too expensive at the volumes Musk is projecting.

Musk is already talking about what comes next. "We are developing exciting chips, including AI6 and Dojo3," he said. He separately provided performance projections for AI6 and AI6.5. Small-batch engineering samples of AI6 are expected in late 2026, with high-volume production targeted for mid-to-late 2027. The Dojo3 reference signals Tesla has not abandoned its troubled Dojo training computer project, despite widespread reports of internal restructuring.

The Musk Performance Claim Problem

The 40x claim deserves scrutiny. Previous Musk hardware announcements have followed a pattern: ambitious performance claims at reveal, followed by quieter revisions during actual production. The original Dojo chip was supposed to revolutionize AI training. It did not. Tesla AI4 was already behind competitors when it shipped. A 40x improvement would be extraordinary for a single generational jump in custom silicon, where 2x to 5x is more typical.

Musk told TSMC and Samsung that "this chip will become one of the highest-volume AI chips in history." That claim depends entirely on whether Tesla ships Optimus and Cybercab at the volumes he has promised. If they do, the AI5 goes into millions of units. If they do not, it is another custom chip that never reaches escape velocity.

What is real: Tesla has a chip design team capable of taking custom silicon from design to tape-out, and two of the world's best foundries are manufacturing it. What is not yet real: the products it is supposed to power at scale.

First reported by TechNode and Tom's Hardware. Tesla has not published official specifications for the AI5.