Overview
- Google announced the TPU 8tx and TPU 8i Max on Monday, Nov. 9, 2026, bringing its eighth‑generation family to four distinct TPU systems launched this year.
- The TPU 8tx provides hardware‑level confidential computing developed with AMD so data and model weights remain encrypted while processing for government and regulated customers.
- The TPU 8i Max is a dual‑chip package that doubles on‑board memory and compute density to support long‑context models that need to hold very large inputs in working memory.
- Google is selling these designs as complete systems that combine silicon, packaging, optical networking and compiler support, and the company is expanding software compatibility with JAX and PyTorch tooling to ease customer adoption.
- The move follows Google’s April split of training and inference silicon into TPU 8t and 8i and signals a commercial push through cloud channels and partnerships, a strategy investors rewarded after the announcement and one that could reshape buyer choices in the AI accelerator market.