Google Unveils Eighth-Generation TPU Chips for the Agentic AI Era
Google introduces two specialized eighth-generation Tensor Processor Units, the TPU 8t for training and TPU 8i for inference, designed to power complex AI agents.
Google unveils its eighth-generation Tensor Processor Units, featuring two specialized chips called the TPU 8t and TPU 8i. These custom processors represent a decade of development and are purpose-built to handle the complex, iterative demands of modern AI agents. By splitting the architecture into two distinct designs, Google aims to optimize both the training and execution phases of artificial intelligence workloads.
The TPU 8t serves as a training powerhouse that speeds up the development of massive and complex AI models. In contrast, the TPU 8i specializes in low-latency inference to support fast, collaborative AI agents that require quick response times. Both chips utilize custom hardware to deliver significant gains in power efficiency and overall performance compared to previous generations.
These new systems are scheduled for general availability later this year to help organizations scale their AI operations. By providing specialized hardware for different stages of the AI lifecycle, Google enables developers to build smarter tools that reason and solve problems more effectively. Customers can currently request more information to prepare for the upcoming release of these advanced supercomputing resources.