---
title: Google unveils two new tensor processing unit systems for artificial intelligence security and computing power
url: https://www.elseif.net/google-unveils-two-new-tensor-processing-unit-systems-for-artificial-intelligence-security-and-computing-power
published: 2026-11-12T16:03:54+00:00
language: en
section: Models
source: https://aitnews.com/2026/11/09/google-tpu-8tx-8i-max-ai-systems/
organizations: Google, AMD
publisher: elseif
---

# Google unveils two new tensor processing unit systems for artificial intelligence security and computing power

Google announced the expansion of its eighth generation tensor processing unit systems with two new configurations: TPU 8tx and TPU 8i Max. The TPU 8tx targets protection of sensitive data and model weights during processing, while TPU 8i Max focuses on increased computing capacity and memory bandwidth for artificial intelligence applications requiring large-scale information processing.

Both systems represent new configurations built on existing eighth generation chips rather than entirely new chip designs. Google stated that system development encompasses integration of computing, memory, networking, packaging, and security technologies.

The TPU 8tx system was designed to run inference applications in confidential computing environments, aiming to protect sensitive data and model weights during use, storage, and transfer. The system targets government organizations and companies in sectors with strict regulatory requirements that need to run proprietary models or applications based on sensitive information. TPU 8tx uses the same chips as TPU 8t systems and employs AMD EPYC fifth generation server processors for task coordination, data preparation, and communication management with artificial intelligence acceleration units. According to AMD, this architecture combines secure encryption technologies for virtual environments and mechanisms for verifying accelerator identity and security settings to help protect information during processing and transfer between components.

Google stated that its Boardfly low-latency network achieves up to a twofold reduction in response time compared to TPU 8t systems with limited performance impact according to company measurements. The system enables model operation and customization within a single architecture, including fine-tuning and reinforcement learning.

TPU 8i Max targets artificial intelligence workloads requiring long contexts and substantial memory, including applications for autonomous agents handling large code repositories or extended visual content. The system combines two TPU 8i units in a single package. Google stated this architecture doubles computing capacity, SRAM memory, and HBM memory bandwidth compared to the base TPU 8i system, helping increase computing density within data center racks.

Google introduced TPU 8t for large-scale training and TPU 8i for low-latency inference in April 2026. With the addition of TPU 8tx and TPU 8i Max, the family now comprises four systems targeting different needs including training, fast inference, confidential computing, and long context processing. Google is also facilitating application development for its processors through support for open source software frameworks including PyTorch and the TorchTPU project.