Tesla's robotaxi is four months from launch, and it will run on last year's chip.
The Cybercab is scheduled to begin production in Q2 2026, Tesla has confirmed, and it will use the current-generation AI4 processor, not the AI5 chip the company just finished designing, according to Electrek. The tape-out, the step where a chip design is sent to manufacturing, was announced April 15, 2026, per Electrek. Volume production of AI5 for vehicles is now targeted for mid-2027, also Electrek. The original promise was that AI5 would be in vehicles in the second half of 2025. That is nearly two years of slip.
The tape-out itself is real. It is an engineering milestone. A tape-out means the design works on paper and has been handed to foundries, Samsung in Taylor, Texas and TSMC in Arizona, to begin turning a blueprint into physical silicon, according to TrendForce. Samples are expected in late 2026, which lines up with Tweaktown's report that high-volume production could begin in late 2026 or early 2027. Volume production in vehicles, however, is still roughly a year away.
Once AI5 does arrive, the performance claims are significant. One AI5 chip delivers compute comparable to NVIDIA's Hopper architecture, while a two-chip configuration approaches Blackwell-level performance at lower cost and power, according to TrendForce. Startup Fortune separately reports Musk's claim that AI5 delivers five times the compute of a dual-AI4 configuration and roughly three times the power efficiency of Blackwell at under 10% of the cost. These are company-reported figures without independent verification, but the gap between Blackwell-level performance and the cost and power envelope Tesla is claiming would be commercially meaningful if it holds.
The dual-sourcing arrangement matters too. Splitting production between Samsung's Taylor facility and TSMC's Arizona operation reduces the risk that any single fab problem, a yield issue or a capacity crunch, delays the entire program. It also signals that Tesla is treating chip supply the same way it treats cell sourcing: not dependent on one vendor.
What the tape-out announcement obscures is AI6. The follow-on chip, targeted for tape-out in December 2026, has already slipped roughly six months because of yield problems at Samsung's 2-nanometer node, pushing mass production to Q4 2027 at the earliest, according to Electrek. The 2nm process is the cutting edge, and it is also the edge where chips break. That AI6 is slipping before tape-out is a bad sign for a company whose autonomous driving program depends on a rolling sequence of next-generation silicon.
The Cybercab launches in three months on hardware that is already one generation behind the chip sitting in a foundry queue. The gap between the promise of AI5 and its arrival in a vehicle is the most honest summary of where Tesla's autonomous driving program stands right now.
