Google announces 8th-generation TPU AI processing chips, including training-focused "TPU 8t" and inference-focused "TPU 8i," with performance per watt doubled

Google announced its 8th-generation TPU lineup, including TPU 8t for training and TPU 8i for inference.

TPU 8t focuses on large-scale parallel training and reliability, while TPU 8i targets lower latency and higher bandwidth.

Both chips deliver 2x better performance per watt than the previous generation and are slated for general availability in late 2026.

The split training/inference design could significantly improve efficiency for building and running large-scale AI systems.