Google adds secure training and higher-memory chips to its TPU range

Google has unveiled two new tensor processing units (TPUs) for AI workloads, expanding its eighth-generation chip range with options for confidential computing and larger memory capacity. The launch supports Google’s effort to rely less on chips from other companies for its cloud business, as major technology firms develop their own AI hardware.

The TPU 8tx, designed with AMD, is built for confidentially training and running AI models, with hardware protections for sensitive data and model weights. Google says it can also run reasoning alongside training, including distilling smaller models using proprietary data. The TPU 8i Max combines two TPU 8i chips in one package, doubling memory, processing power and bandwidth to help models handle more information during queries.