Our eighth generation TPUs: two chips for the agentic era
Summary
Google announces the eighth generation of its Tensor Processor Units, TPU 8t for training and TPU 8i for low latency inference. The two chips are designed for agentic workloads with co designed silicon, networking and software, and push scale, efficiency and performance for frontier AI. The rollout aims to support large scale training and autonomous agent workloads, with availability later this year as part of Google Cloud AI Hypercomputer.