Google unveils two new TPUs designed for the "agentic era"

Google unveiled two eighth-generation TPU designs: TPU 8t for training and TPU 8i for inference.

The new chips improve memory handling, fault tolerance, telemetry, larger TPU pods, and on-chip SRAM.

Google also added an Axion ARM-based host setup to boost overall efficiency for large-scale AI workloads.

The TPUs are meant to cut the cost and waste of running large models, making agentic AI systems more practical at scale.