Google has officially launched its eighth-generation AI chips, TPU8t and TPU8i, at the Google Cloud Next 2026 conference. According to Jin10, this marks the first time Google has designed separate chips for training and inference since the inception of TPU. The new chips feature updated configuration schemes for cluster networking. China International Capital Corporation (CICC) suggests that this specialized AI hardware infrastructure is expected to enhance the efficiency of large model training and inference, thereby reinforcing the 'flywheel effect' of cost reduction and increased demand. From an industry chain perspective, CICC remains optimistic about core investment areas such as optical communication, chip customization services, storage, and wafer foundry.