Google challenges Nvidia with new dedicated training and inference AI chips

The TPU 8t targets model training while the TPU 8i handles inference workloads, a split from previous generations that handled both tasks