Google presents TPU 8t and TPU 8i chips; splits training and inference

At Cloud Next 2026, Google unveiled its eighth-generation TPUs as two separate chips: TPU 8t and TPU 8i.

TPU 8t is built for large-scale model training, with major gains in compute, memory, and cluster scale.

TPU 8i targets inference workloads, emphasizing lower latency, more memory per pod, and higher chip density.

The split product strategy is meant to sharpen Google’s cloud AI stack and compete more directly with Nvidia.