Google Expands TPU Line With Confidential Computing and High-Memory AI Chips

Google has added two new variants to its tensor processing unit (TPU) family, extending its in-house AI chips to confidential inference and larger-memory workloads as it seeks to reduce reliance on third-party chips for its cloud business.

The TPU 8tx and TPU 8i Max belong to the eighth generation of TPUs that Google introduced in April, which focused on greater inference capacity and support for large training workloads.

Designed with AMD and based on the TPU 8t, the TPU 8tx is built specifically for confidentially training and running AI models. Google said the chip embeds security at the hardware level, keeping sensitive data and model weights protected and isolated even while in use. A custom inferencing network allows it to run reasoning alongside training, enabling companies to distill and train small models on proprietary data without moving it between devices.

The TPU 8i Max combines two inference-focused TPU 8i chips in a single package, doubling memory capacity, processing power and bandwidth so AI models can handle a wider range of information during queries.

The move comes as Microsoft, Amazon, OpenAI and Anthropic develop their own AI chips to lessen dependence on Nvidia, whose CEO Jensen Huang admitted over the weekend to an affair with someone alleged to be a Chinese spy.