TPU 8tx and TPU 8i Max

Regulated businesses can now perform AI inference and model fine-tuning on their most sensitive data without leaving a secure enclave.

Google has expanded its eighth-generation Tensor Processing Unit (TPU) portfolio with two new systems: TPU 8tx and TPU 8i Max. The TPU 8tx is designed for secure, confidential inference, offering hardware-level protection for sensitive data and model weights during active use, and is suitable for regulated industries and governments. It also supports small-scale training and fine-tuning of models on private data without data migration. The TPU 8i Max is built for agentic workloads that require massive context windows and high compute density per rack. These additions join the previously announced TPU 8t for large-scale training and TPU 8i for real-time reasoning, marking the introduction of four new TPU systems within a single year. Google aims to make these TPUs more accessible through PyTorch support and open-source software, enabling developers to build and deploy AI models more efficiently.

The introduction of TPU 8tx and TPU 8i Max signifies Google's strategic response to the evolving demands of the "agentic era," where AI systems increasingly require specialized hardware for security, massive context handling, and efficient reasoning. TPU 8tx's hardware-level confidential computing capabilities directly address a critical need for regulated industries and enterprises handling proprietary data, enabling secure inference and model fine-tuning without data exposure. This focus on security-integrated hardware is a key differentiator as AI workloads become more complex and sensitive.

The TPU 8i Max caters to the growing requirement for handling extremely large context windows, a crucial factor for advanced AI agents and reasoning tasks. By expanding its eighth-generation TPU offerings to four distinct systems—8t for training, 8i for reasoning, 8tx for secure inference, and 8i Max for large contexts—Google is providing a comprehensive, specialized hardware suite for the entire AI lifecycle. This modular, system-level approach, coupled with an emphasis on software accessibility via PyTorch, allows Google to rapidly deliver tailored solutions to meet the fast-paced needs of the AI industry.