VMTech
Discuss a project →

Google Introduces TPU 8tx and TPU 8i Max AI Chips

Google Introduces TPU 8tx and TPU 8i Max AI Chips

Google has introduced two new additions to its eighth-generation Tensor Processing Unit portfolio: TPU 8tx and TPU 8i Max. The company says the chips extend the use cases addressed by the TPU 8t and TPU 8i, launched in April, with TPU 8tx aimed at confidential AI training and inference and TPU 8i Max designed for larger, memory-intensive inference workloads.

The launch is part of Google’s effort to expand its in-house AI hardware capabilities for cloud computing, reducing dependence on chips supplied by other companies. Google developed TPU 8tx with AMD and based it on the TPU 8t design, while TPU 8i Max combines two inference-focused TPU 8i chips in a single package.

Confidential operations at the hardware layer

Google describes TPU 8tx as a chip for training and running AI models confidentially. It builds security into the hardware layer so that sensitive data and model weights remain protected and isolated while they are actively being used. The intended use is particularly relevant to organisations that need to process proprietary information without moving it between devices.

The chip also includes a custom inference network that can run reasoning alongside training. Google says this enables the distillation and training of smaller models on the fly. That capability can support workflows in which a company trains a model on its own data while keeping that data within the relevant environment.

More capacity for inference workloads

TPU 8i Max is focused on inference rather than confidential computing. By bringing together two TPU 8i chips, Google says it doubles memory capacity, processing power and bandwidth in one package. The company positions the resulting configuration for AI models that need to handle a larger range of information during queries.

The new variants build on the eighth-generation TPU family, whose initial models brought increased inference capacity and support for large training requirements. The foundations of this platform were set by the TPU 8t and TPU 8i releases with the TPU 8t and TPU 8i releases, and the new variants narrow their focus toward protected workloads and higher-capacity inference.

Part of a broader custom-chip push

Google’s announcement arrives as other hyperscalers and AI labs develop their own AI-specific semiconductors and associated software. Microsoft, Amazon, OpenAI and Anthropic are all cited among the organisations pursuing alternatives that can reduce reliance on Nvidia’s dominant AI hardware position.

For businesses using cloud AI services, the practical implication is that workload design increasingly matters when selecting infrastructure. Teams handling sensitive proprietary data can evaluate confidential-computing options, while teams with memory-heavy query workloads can assess whether higher-capacity inference hardware better matches their model requirements.

#aichips#googlecloud#confidentialai#cloudinfrastructure
Open analytics
On the site 6 views
min read 3 10.11.2026
Instagram

Google Introduces TPU 8tx and TPU 8i Max AI Chips

Open the post on Instagram ↗