Google has unveiled its next-generation Tensor Processing Unit (TPU), a cutting-edge chip designed for artificial intelligence (AI) training and powering AI agents.
At its annual cloud computing conference in Las Vegas on April 22, Google showcased the high-performance TPU as a cornerstone of its innovative technology.
The TPU is a specialized semiconductor engineered to rapidly process large-scale numerical data (tensors) that AI systems use for learning and decision-making. With this development, Google aims to reduce its dependence on NVIDIA’s graphics processing units (GPUs), which currently dominate the market alongside Amazon.
Among the new TPUs, one is tailored for training large language models that drive AI, while the other is dedicated to the reasoning and decision-making processes employed by AI agents. These AI agents can function as sophisticated digital assistants, capable of making independent judgments and executing complex, multi-step tasks when given a specific goal.
Thomas Kurian, Chief Executive Officer (CEO) of Google Cloud, announced that the TPU, developed in partnership with semiconductor giant Broadcom, is slated for release by the end of this year.
In a blog post, Google CEO Sundar Pichai emphasized the need to advance infrastructure to handle the most demanding AI workloads in the era of AI agents. He revealed that this year, Google will introduce its 8th generation TPU, featuring an innovative dual-chip design.