Google has unveiled a new generation of Tensor Processing Units (TPUs) designed to train artificial intelligence models and power the fast-growing wave of digital “agents.”
The announcement came at the company’s annual cloud computing conference in Las Vegas on Wednesday, where high-performance AI infrastructure took centre stage.
Like Amazon, Google has increasingly moved to design its own advanced chips, aiming to reduce reliance on Nvidia’s dominant graphics processing units (GPUs).
“In the era of AI agents, infrastructure needs to evolve to handle the most demanding workloads,” chief executive Sundar Pichai said in a blog post. “This year, we’re introducing the eighth generation of our Tensor Processing Units with a dual-chip approach.”
One of the new TPUs is optimised for training large language models, while the other is tailored for “inference” — the reasoning and decision-making processes that underpin AI agents.
AI agents are autonomous digital assistants capable of carrying out complex computing tasks with minimal human input.
Developed in partnership with Broadcom, the chips are expected to become available later this year, according to Thomas Kurian.
The launch comes amid intensifying competition in the AI hardware space. Earlier this year, Nvidia announced its next-generation Vera and Rubin GPUs, while Amazon introduced an updated version of its custom Trainium processors.
Despite the push toward in-house silicon, Google, Amazon and Microsoft continue to rely heavily on Nvidia GPUs as part of their cloud infrastructure.
AFP
