Google launches TPU 8 chips with 3x power to speed AI training, cut cloud costs

Google unveiled TPU 8t for large-model training and TPU 8i for inference and AI agents at Cloud Next.

Both chips will be available on Google Cloud later this year and are set to anchor the AI Hypercomputer stack.

TPU 8t is aimed at improving training efficiency and lowering costs, while TPU 8i targets faster, more efficient inference.

The launch strengthens Google Cloud’s pitch as a lower-cost alternative to Nvidia-based AI infrastructure.