Overview
- Elon Musk confirmed AI5 will be fabricated at Samsung’s Texas plant and TSMC’s Arizona facility, revising an earlier plan that centered on a single foundry.
- Musk said Tesla will intentionally overproduce AI5 so unused chips can be deployed in vehicles, the Optimus robot program, or the company’s data centers.
- He claimed AI5 could reach up to 40x the performance of AI4 and said Tesla removed legacy GPU and image signal processor blocks, enabling a half‑reticle design.
- Tesla will continue to rely on Nvidia for training, reporting compute equivalent to 81,000 H100 chips, and plans to use AI5 in combination with Nvidia systems.
- Samsung’s role expands alongside a confirmed $16.5 billion AI6 contract, with reports pointing to 2026 AI5 mass production and a broader foundry rebound, and Musk said Samsung’s Texas tools are slightly more advanced than TSMC’s Arizona line.