On April 30th, CICC released a research report stating that at the Google Cloud Next 2026 conference, Google officially unveiled its eighth-generation in-house AI chips, TPU8t and TPU8i. This launch marks Google's first separation of training chips from inference chips in design, and introduces updated configurations for cluster networking. CICC believes that this leanly divided AI hardware infrastructure is expected to enhance the training and inference efficiency of large models, further driving a virtuous cycle of 'supply cost reduction—demand increase.' From an industrial chain perspective, core investment directions such as optical communication, chip customization services, storage, and wafer foundry are worth noting.
