Google launches two dedicated AI chips, expanding confidential computing and memory capacity.

Google unveiled two eighth-generation tensor processing unit variants, the TPU 8tx and TPU 8i Max, expanding its custom AI chip lineup. The TPU 8tx, co-designed with AMD and built on the TPU 8t, targets confidential model training and inference with hardware-level protection for in-use data and model weights, plus a custom inference network that runs inference during training for instant distillation and small-model training. The TPU 8i Max packs two inference-focused TPU 8i chips into a single package, doubling memory capacity, processing power and bandwidth so models can handle larger information ranges at query time. The launch matters because it deepens Google's vertical integration in AI silicon and pushes confidential computing into training workloads, while the AMD co-design signals a broadening partner ecosystem. Next, watch for pricing and cloud availability of both chips and whether the doubled-memory 8i Max shifts the cost calculus for large-scale inference.