Tesla has updated its AI accelerator roadmap, bringing its chips under a unified SI (Super Intelligence) series naming convention, with the original HW4/AI4 now designated as SI4. The sixth-generation inference accelerator, SI6, is expected to launch between 2029 and 2030. It will feature a 2nm-class process, multi-chip module packaging, 144GB of memory, and a compute target 80 times that of SI4.
