Google Cloud officially introduced the eighth generation of Tensor Processing Units (TPU), namely TPU 8t and TPU 8i which are designed to accelerate artificial intelligence (AI) workloads in the Agentic era .
SVP and Chief Technologist , AI and Infrastructure, Google Cloud, Amin Vahdat at The Future of AI Infrastructure, Google Cloud Next 26, in Las Vegas, United States, Tuesday, said that this year, two specialized chips were launched simultaneously, to welcome the era of Agentic AI .
"For the first time, two purpose-built TPUs, the 8t and 8i, were completely designed from the ground up to meet today's needs. The 8t TPU is for training . It's the main engine. It's the largest scale. The 8i TPU is truly built for inference ," Vahdat said.
Vahdat stated that TPU has supported leading baseline models, including Gemini, for years. This eighth-generation TPU integration will deliver scale, efficiency, and capabilities spanning training, serving, and agentic workloads .
