PyTorch Integration with TPU

Google Launches 8th‑Gen TPUs Split for Training and Inference

The move targets lower inference costs for large agentic AI workloads.