Google expands TPU generation: 8tx with AMD EPYC for confidential AI and 8i Max with double memory bandwidth - igor´sLAB

While NVIDIA and AMD compete for data center operators with increasingly powerful AI accelerators, Google is pursuing its own path with specialized solutions. The TPU 8tx is designed to protect sensitive data and AI models during processing, while the TPU 8i Max targets applications requiring large context windows and high memory bandwidth.

Google had previously announced the base variants TPU 8t and TPU 8i in April 2026, with the former focused on training large AI models and the latter on inference tasks with low response times. The new additions expand the family to four systems total.

Enhanced Security Features

The TPU 8tx uses the same silicon as the TPU 8t but incorporates a modified system architecture featuring fifth-generation AMD EPYC processors and Google's Boardfly network technology. According to Google, this combination reduces latency by up to 50 percent compared to the TPU 8t while adding security features with minimal performance impact.

Google targets the TPU 8tx at companies and government agencies handling confidential information, including financial services providers, public administrations, and organizations with proprietary AI models. The system implements confidential computing to better protect data during processing through isolated execution environments.

Memory Bandwidth Focus

The TPU 8i Max takes a different approach by combining two TPU 8i chips within a single package. This configuration doubles compute performance, SRAM capacity, and HBM memory bandwidth compared to a single TPU 8i, addressing the growing needs of memory-intensive AI applications.

Modern AI applications increasingly process extensive documents, large software projects, or longer video sequences, often with multiple AI agents working in parallel on related tasks. These demands make memory bandwidth and local capacity critical performance factors that Google aims to address with the TPU 8i Max.

Market Strategy

Google's strategy focuses on closely integrated platforms combining its own accelerators, networking technology, and cloud infrastructure, rather than comprehensive GPU platforms like NVIDIA. The company reports that nearly 100 customers now use TPU platforms, including commercial companies and research institutions.

The new systems demonstrate how competition in AI is shifting from individual processors to fully integrated platforms tailored to specific requirements. However, Google has not yet announced specific prices, delivery dates, or availability for the TPU 8tx and 8i Max, and independent performance comparisons with current NVIDIA and AMD platforms remain unavailable.