Google has introduced two new AI processing systems, TPU 8tx and TPU 8i Max, designed to improve data security and support larger artificial intelligence workloads. Announced on November 9, 2026, the additions expand Google’s eighth-generation Tensor Processing Unit (TPU) portfolio as the company strengthens its custom semiconductor offerings for cloud computing.
According to Google’s official announcement, TPU 8tx focuses on confidential computing, while TPU 8i Max provides greater memory capacity and processing performance for complex AI applications. The systems join TPU 8t and TPU 8i, introduced in April 2026, bringing the company’s eighth-generation lineup to four configurations.
TPU 8tx Introduces Hardware-Level Security for AI Workloads
TPU 8tx is designed for organizations that need to train and operate AI models using confidential information. Developed using Google’s existing TPU 8t architecture and fifth-generation AMD EPYC processors, the system incorporates hardware-level protections for sensitive data and model parameters.
Google said the technology keeps information isolated and protected while it is actively being processed, addressing security concerns associated with traditional computing environments. The capability is particularly relevant to regulated industries and organizations working with proprietary datasets.
The system also incorporates a low-latency network that supports AI reasoning alongside model training. According to Google, TPU 8tx can deliver up to twice the latency performance of TPU 8t while maintaining similar computing efficiency.
The configuration allows developers to fine-tune existing models, train smaller models using larger systems, and perform reinforcement learning without transferring sensitive data between separate computing clusters.
TPU 8i Max Doubles Memory and Processing Capacity
The second system, TPU 8i Max, combines two TPU 8i processors into a single package to support AI applications requiring substantial memory resources. Google said the design doubles computing performance, on-chip SRAM capacity, and high-bandwidth memory throughput compared with a single TPU 8i.
The additional capacity is intended to support AI models processing large amounts of information, such as extensive software repositories, lengthy videos, and complex requests involving multiple AI agents. These workloads can require substantial memory bandwidth to maintain performance.
TPU 8i Max also increases computing density by fitting twice the processing silicon within an existing rack footprint. This allows data centers to support larger AI workloads without requiring proportional increases in physical space.
Google Expands Software Support for Custom AI Chips
Alongside the hardware announcement, Google Cloud is expanding software support to make TPUs easier for developers to use. Recent developments include support for PyTorch, Hugging Face models, and open-source inference frameworks such as vLLM and SGLang.
Google also introduced tools for automating TPU setup and optimization through AI coding assistants. These updates are intended to reduce the technical effort required to adapt existing AI applications to Google’s hardware.
The expansion comes as Microsoft, Amazon, OpenAI, and Anthropic develop custom AI chips to reduce reliance on Nvidia’s processors. Google has not disclosed pricing or specific availability dates for the two new TPU configurations.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.