Google unveils TPU 8i Max and 8tx — accelerators for AI agents

On November 9, 2026, Google expanded its eighth-generation TPU lineup to four systems, adding the TPU 8tx with hardware confidential computing and the dual-chip TPU 8i Max, while simultaneously open-sourcing TorchTPU, XLA Serving Engine, and TPU Sync.

Compiled by Sergey KostenchukPublished 2026-11-09Updated 2026-11-10

2026-11-09 Coding Google

💻 Google reveals TPU 8i Max and 8tx — chips for AI agents

Google has expanded its eighth-generation TPU lineup to four systems: 8tx — TPU 8t silicon with hardware confidential computing; 8i Max — two 8i chips in a single enclosure, doubling power, SRAM, and HBM. TorchTPU, XLA Serving Engine, and TPU Sync have been open-sourced.

🌍 Google has released four TPU systems for the first time in a year and transitioned to ready-made platforms — NVIDIA, AMD, and AWS Trainium now must compete not only on performance but also on data protection.

👤 TorchTPU runs PyTorch and Hugging Face models on TPU without rewriting. 8tx is for scenarios where data cannot be decrypted in memory, while 8i Max is for contexts of millions of tokens. Prices and GA dates are unknown.

Source 1: https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/tpu-8i-max-and-8tx/ Source 2: https://www.datacenterdynamics.com/en/news/google-expands-eighth-generation-tpu-portfolio-with-two-new-tpu-systems/

Sources