# Google adds TPU 8tx and TPU 8i Max to its eighth-generation chips

> One chip is for confidential inference, the other doubles compute and memory per package. Google has given no price or release date.

URL: https://claudegpt.si/news/google-tpu-8tx-8i-max  
Published: 2026-11-09T14:53:41.063Z  
Author: ClaudeGPT Si Newsroom  
Section: AI News  
Topics: Google, AI chips, Google Cloud, AI infrastructure  
Publisher: ClaudeGPT Si

## Key takeaways

- Google announced the TPU 8tx and TPU 8i Max on November 9, its third and fourth TPU systems of the year.
- The TPU 8tx keeps data and model weights protected while in use, and the TPU 8i Max doubles compute, SRAM and memory bandwidth per package versus the TPU 8i, according to Google.
- Google gave no price, release date or absolute specs.

Google announced two more chips in its eighth-generation TPU family on November 9: the TPU 8tx, built for confidential inference, and the TPU 8i Max, a double-chip package for very long context windows. [Google's own post](https://cloud.google.com/blog/topics/ai-infrastructure/tpu-8tx-and-tpu-8i-max), written by Amin Vahdat, says this is the first time the company has introduced four TPU systems in one year. A TPU is Google's own AI accelerator, the chip it offers as an alternative to Nvidia's GPUs.

The TPU 8tx runs on the same silicon as the TPU 8t, which Google introduced in April for training, and is paired with custom AMD CPU hosts. Google says its hardware keeps sensitive data and model weights protected and isolated even while a model is using them, which is aimed at regulated companies and governments. Teams can fine-tune on private data, distill a large model into a smaller one and train agents with reinforcement learning on one system, without moving data between clusters. Google says its Boardfly network gives up to two times lower latency than the TPU 8t, with a near-zero performance penalty.

The TPU 8i Max puts two TPU 8i chips on one package. Compared with the TPU 8i, Google says it has double the compute, double the SRAM and double the high-bandwidth memory bandwidth, and fits twice the silicon in the same rack space. It is meant for large context windows and high compute density per rack.

The post gives no release date, no price and no absolute figures such as FLOPS or memory size. Every comparison is relative to another Google chip, and all the numbers are Google's own, not independent tests. Google also says nearly 100 customers now use TPUs, including frontier labs, trading firms and retailers, and that TorchTPU, which it open-sourced in October, lets PyTorch code and Hugging Face models run natively on TPUs.

For people choosing AI hardware, the confidential option is the part to watch. It targets companies that cannot send private data to a shared cluster, and Google has not yet said when they can buy it.

## Sources

- [Our newest TPUs: 8tx and 8i Max, built for the agentic era](https://cloud.google.com/blog/topics/ai-infrastructure/tpu-8tx-and-tpu-8i-max), Google Cloud
