Tesla unveils AI5 processor sample with claimed 40X performance boost over predecessor

Elon Musk has showcased the first sample of Tesla's AI5 processor, which the company says delivers up to 40 times the performance of its AI4 predecessor in certain scenarios. The AI5 module features a relatively small application-specific integrated circuit (ASIC) die surrounded by 12 memory packages from SK hynix, likely using GDDR6 or GDDR7 memory technology. Tesla's chip design team recently completed the "tape-out" process, marking the final design stage before production. Musk stated the AI5 processor will be used to power AI applications in Tesla vehicles, Optimus robots, and potentially xAI data centers. He also mentioned that next-generation AI6 and Dojo 3 processors are currently in development. The module uses an organic substrate and industry-standard memory packages, with 12 memory packages suggesting a 384-bit memory interface. This configuration could provide memory bandwidth between 768 GB/s and 1.536 TB/s depending on the memory type employed. During Tesla's third-quarter 2025 earnings call, Musk emphasized the chip's efficiency, noting that the AI5 design fits on half a reticle with adequate margin for memory traces, CPU cores, and PCIe blocks. The sample shown features "KR 2613" markings, indicating packaging during the 13th week of 2026. Tesla plans to manufacture the AI5 through both Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung Foundry, though the specific foundry for the current sample remains undisclosed. Assuming no design revisions are required, deployment could begin in 2027. The announcement also confirms Tesla continues development of its Dojo system-on-wafer processor for AI training, despite previous reports suggesting the project had been abandoned. Musk has indicated that AI6 and Dojo 3 may feature a converged architecture that would unify the company's software and hardware stacks.